Add rating snapshots, percentile, and exam readiness to the dashboard API

Backend half of the dashboard stat-card redesign. Three new data
points the old DashboardResponse had no way to compute:

- Week-over-week rating deltas: UserRating/CcsUserRating only ever
  stored the current Elo (mutated in place, UpdatedUtc just means
  "last touched"), so there was no way to know what a rating was 7
  days ago. Added UserRatingSnapshot (new collection, unique on
  UserId+DateUtc) and UserRatingSnapshotService, a BackgroundService
  following AnswerStatisticsMaintenanceService's exact shape (a
  PeriodicTimer plus a Mongo lease so replicas behind the HPA don't
  race), writing one snapshot per user per day. No historical
  backfill - deltas are simply null for a user's first ~7 days after
  this ships, which the DTO represents honestly via nullable ints
  rather than a fake "+0".
- Percentile rank: DashboardStore now counts how many users have a
  higher OverallElo/Elo than the current user out of the total.
- Exam Readiness + projected pass rate: reuses the existing, tested
  IRT-based ExamSimulationScore.PassProbability (a real normal-CDF
  estimate of scoring >=200, computed in ExamSimulationScoring.Calculate)
  rather than inventing a new prediction model. DashboardStore pulls
  the latest completed simulation's score directly and maps
  PassProbability to a High/Medium/Low label via tunable thresholds.
- Average Score's "+6% vs last 100" needed no new data source at all -
  just one more slice (Skip(100).Take(100)) over the QuestionAttempts
  already loaded for FirstAttemptAccuracy.

Added DashboardStore.Percentile/ReadinessLabel as pure static helpers
(matching the existing Percentage/CalculateCurrentStreak pattern) and
unit tests for both. Full backend suite: 165 passed, 0 failed.

api/QuestionsApi/Services/CcsRatingStore.cs

@@ -31,6 +31,12 @@ public sealed class CcsRatingStore
         return new CcsRatingProfileResponse { Elo = rating.Elo, RatedCaseCount = rating.RatedCaseCount };
     }
 
+    public async Task<CcsUserRating> GetOrInitializeHistoryAsync(string userId, DateTime nowUtc)
+    {
+        var rating = await GetOrCreateUserAsync(userId);
+        return await EnsureHistoryBaselineAsync(rating, nowUtc);
+    }
+
     public async Task<CcsRatingChange> ApplyAsync(CcsAttempt attempt)
     {
         var actual = Math.Clamp((attempt.Score?.Percent ?? 0) / 100.0, 0, 1);
@@ -52,6 +58,7 @@ public sealed class CcsRatingStore
         }
 
         var user = await GetOrCreateUserAsync(attempt.UserId);
+        user = await EnsureHistoryBaselineAsync(user, DateTime.UtcNow);
         var caseRating = await GetOrCreateCaseAsync(attempt.CaseExternalId);
         var expected = ExpectedScore(user.Elo, caseRating.Elo);
         var userDelta = (int)Math.Round(UserK(user.RatedCaseCount) * (actual - expected), MidpointRounding.AwayFromZero);
@@ -137,6 +144,26 @@ public sealed class CcsRatingStore
         }
     }
 
+    private async Task<CcsUserRating> EnsureHistoryBaselineAsync(CcsUserRating rating, DateTime nowUtc)
+    {
+        if (rating.HistoryBaselineElo.HasValue && rating.HistoryStartedUtc.HasValue)
+            return rating;
+
+        var result = await _users.UpdateOneAsync(
+            value => value.UserId == rating.UserId && value.HistoryStartedUtc == null,
+            Builders<CcsUserRating>.Update
+                .Set(value => value.HistoryBaselineElo, rating.Elo)
+                .Set(value => value.HistoryStartedUtc, nowUtc));
+        if (result.ModifiedCount > 0)
+        {
+            rating.HistoryBaselineElo = rating.Elo;
+            rating.HistoryStartedUtc = nowUtc;
+            return rating;
+        }
+
+        return await _users.Find(value => value.UserId == rating.UserId).FirstAsync();
+    }
+
     private async Task<CcsCaseRating> UpdateCaseAsync(CcsCaseRating current, int delta)
     {
         while (true)