feat(job-search): extract keywords from LLM match call instead of heuristics

Piggybacks keyword extraction onto the existing CV-to-job LLM call —
no extra API calls. The system prompt now instructs the model to return
8-12 English job-search terms (job titles, technologies, skills, domains)
in a new `keywords` field alongside the existing score/summary fields.

Keywords flow: LLM JSON → JobMatchResponse.Keywords → CreateJobSearchTokenRequest →
JobSearchTokenEntity.Keywords (stored comma-separated) → JobSearchSessionEntity.Keywords
(copied at session-creation time, no RAG call needed).

Changes:
- Add Keywords to JobMatchResponse, CreateJobSearchTokenRequest, JobSearchTokenEntity
- IJobTokenService.CreateTokenAsync now accepts IReadOnlyList<string> keywords
- JobTokenService: store keywords on token; TriggerStartAsync reads token.Keywords
  instead of fetching CV text from RAG — removes IRagApiClient dependency
- Remove heuristic ExtractKeywords method
- Migration AddKeywordsToJobSearchTokens: adds Keywords column to cvSearch.JobSearchTokens
- Migration UpdateCvMatchSystemPromptKeywords: updates ai.cv-match.system-prompt seed
  to include keywords in the JSON shape

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Apis/cv-matcher-api/Services/Contracts/IJobTokenService.cs

@@ -12,12 +12,13 @@ public interface IJobTokenService
     /// <param name="cvDocumentId">Identifier of the indexed CV document.</param>
     /// <param name="email">Email address of the user who will receive the results.</param>
     /// <param name="language">Preferred language for result emails (e.g. <c>"en"</c>, <c>"ro"</c>).</param>
+    /// <param name="keywords">Job search keywords extracted by the LLM during the match call.</param>
     /// <param name="ct">Cancellation token.</param>
     /// <returns>
     /// The generated token ID to embed in the one-click job search link,
     /// or <c>null</c> when no job providers are currently enabled (link should be suppressed).
     /// </returns>
-    Task<string?> CreateTokenAsync(string cvDocumentId, string email, string language, CancellationToken ct);
+    Task<string?> CreateTokenAsync(string cvDocumentId, string email, string language, IReadOnlyList<string> keywords, CancellationToken ct);
 
     /// <summary>
     /// Validates the token and, if valid, marks it as used and creates a <c>Pending</c> job search session.

Apis/cv-matcher-api/Services/JobTokenService.cs

@@ -1,6 +1,4 @@
 using System.Text.Json;
-using System.Text.RegularExpressions;
-using Api.Clients.Api.Contracts;
 using Api.Services.Contracts;
 using CvMatcher.Models.Responses;
 using CvSearch.Data;
@@ -16,28 +14,27 @@ namespace Api.Services;
 /// Provider configuration is read from <c>cvSearch.JobProviders</c> at session-creation time and
 /// snapshotted into <c>JobSearchSessionEntity.ProviderConfigJson</c> so subsequent config changes
 /// do not affect already-queued sessions.
+/// Keywords are extracted by the LLM during the CV-to-job match call and stored on the token,
+/// then copied to the session when the user clicks the link — no extra RAG call needed.
 /// </summary>
 public sealed class JobTokenService : IJobTokenService
 {
     private readonly CvSearchDbContext _db;
-    private readonly IRagApiClient _rag;
     private readonly JobSearchSettings _settings;
     private readonly ILogger<JobTokenService> _logger;
 
     public JobTokenService(
         CvSearchDbContext db,
-        IRagApiClient rag,
         IOptions<JobSearchSettings> settings,
         ILogger<JobTokenService> logger)
     {
         _db = db;
-        _rag = rag;
         _settings = settings.Value;
         _logger = logger;
     }
 
     /// <inheritdoc />
-    public async Task<string?> CreateTokenAsync(string cvDocumentId, string email, string language, CancellationToken ct)
+    public async Task<string?> CreateTokenAsync(string cvDocumentId, string email, string language, IReadOnlyList<string> keywords, CancellationToken ct)
     {
         var hasEnabledProviders = await _db.JobProviders.AnyAsync(p => p.Enabled, ct);
         if (!hasEnabledProviders)
@@ -52,6 +49,7 @@ public sealed class JobTokenService : IJobTokenService
             CvDocumentId = cvDocumentId,
             Email = email,
             Language = language,
+            Keywords = string.Join(",", keywords),
             ExpiresAt = DateTime.UtcNow.AddDays(_settings.TokenExpiryDays),
             Used = false,
             CreatedAt = DateTime.UtcNow
@@ -59,7 +57,7 @@ public sealed class JobTokenService : IJobTokenService
 
         _db.JobSearchTokens.Add(token);
         await _db.SaveChangesAsync(ct);
-        _logger.LogInformation("Job search token created. TokenId={TokenId}, CvDocumentId={CvDocumentId}", token.Id, cvDocumentId);
+        _logger.LogInformation("Job search token created. TokenId={TokenId}, CvDocumentId={CvDocumentId}, Keywords={Keywords}", token.Id, cvDocumentId, token.Keywords);
         return token.Id;
     }
 
@@ -74,8 +72,7 @@ public sealed class JobTokenService : IJobTokenService
         token.Used = true;
         await _db.SaveChangesAsync(ct);
 
-        var cv = await _rag.GetDocumentAsync(token.CvDocumentId, ct);
-        var keywords = cv is not null ? ExtractKeywords(cv.Text) : string.Empty;
+        var keywords = token.Keywords;
 
         var enabledProviders = await _db.JobProviders
             .Where(p => p.Enabled)
@@ -108,10 +105,6 @@ public sealed class JobTokenService : IJobTokenService
         return StartJobSearchStatus.Started;
     }
 
-    /// <summary>
-    /// Maps a <see cref="JobProviderEntity"/> to the <see cref="JobProviderConfig"/> DTO used by
-    /// <c>cv-search-job</c>. The <c>InitialKeywords</c> list is stored as a JSON array in the entity.
-    /// </summary>
     private static JobProviderConfig ToConfig(JobProviderEntity entity)
     {
         List<string> keywords;
@@ -136,27 +129,4 @@ public sealed class JobTokenService : IJobTokenService
         };
     }
 
-    /// <summary>
-    /// Extracts up to 10 meaningful keywords from the CV text using simple heuristics (no LLM).
-    /// Samples the first 2000 characters (where title/role/skills usually appear), splits by
-    /// whitespace and common delimiters, strips punctuation, and deduplicates.
-    /// Works regardless of whether the PDF extractor preserves newlines.
-    /// </summary>
-    private static string ExtractKeywords(string cvText)
-    {
-        // Focus on the header area where name/title/skills typically appear
-        var sample = cvText.Length > 2000 ? cvText[..2000] : cvText;
-
-        var words = sample
-            .Split([' ', '\n', '\r', '\t', '|', '/', ',', ';', '(', ')'], StringSplitOptions.RemoveEmptyEntries)
-            .Select(w => Regex.Replace(w, @"[^\w\-]", "").Trim('-'))
-            .Where(w => w.Length > 2)
-            .Where(w => !Regex.IsMatch(w, @"^[\d\-]+$"))  // skip phone fragments and pure numbers
-            .Where(w => !w.Contains('@') && !w.Contains('.'))  // skip emails and URLs
-            .Distinct(StringComparer.OrdinalIgnoreCase)
-            .Take(10)
-            .ToList();
-
-        return string.Join(",", words);
-    }
 }