Implement multi-chapter deduplication algorithm

Implement deduplication as a separate BufdirDeduplicator class rather than embedding it in BufdirActivityQueryService — this keeps concerns separated and makes the algorithm independently testable. The fingerprint key should be: '${record.peerMentorId}|${record.activityType}|${record.activityDate.toIso8601String().substring(0, 10)}|${record.durationMinutes}'. Use a LinkedHashMap to preserve insertion order of surviving records. When a collision is detected, retain the record with the lexicographically smaller chapterId for full determinism.

Output the DeduplicationSummary as a simple value object: {originalCount, deduplicatedCount, removedCount, conflictingChapterPairs: List}. This summary should be logged at debug level and surfaced in the UI export confirmation screen. Be careful with activityDate: normalise to midnight UTC before fingerprinting to handle timezone edge cases where the same activity appears on different calendar dates in UTC vs local time.

Unit tests using flutter_test only — no Supabase dependency needed since deduplication is a pure function. Required test cases: (1) list with no duplicates returns identical list, (2) two records with identical fingerprint → one removed, (3) three records from three chapters with same fingerprint → one remains, (4) two peer mentors with same activity fields but different peerMentorId → both retained, (5) empty list returns empty list, (6) list with 1000+ records (performance test with stopwatch assertion), (7) determinism test: shuffle input and assert same output. Use table-driven test style with a helper that builds BufdirActivityRecord test fixtures. Target 100% branch coverage on the deduplication method.