P4-W2: Export jobs — async dataset export with JSONL and CSV support

[user1303836]

Summary

• Adds async export job system for Silver datasets with 3 new API endpoints:
  • POST /v1/export/dataset — create an export job by dataset name, target, and format (JSONL or CSV)
  • GET /v1/export/jobs/:job_id — poll job status and metadata
  • GET /v1/export/jobs/:job_id/download — download completed export file
• Introduces ExportFormat enum in core with JSONL and CSV variants
• Adds 6 export query methods in adapters (one per Silver dataset)
• Full job lifecycle in api: creation → background processing → download
• Fixed a String::leak() memory leak in the download handler
• Preserves existing GET /v1/export/:wallet wallet export endpoint unchanged
• 24 new tests covering format selection, job lifecycle, dataset routing, and error cases
• README.md and CLAUDE.md route lists updated to reflect new endpoints

Validation

All 3 required checks pass:

• cargo fmt --all --check ✓
• cargo clippy --workspace --all-targets -- -D warnings ✓
• cargo test --workspace — 737 tests pass ✓

No migration changes. No breaking API changes. No wallet-assumption regressions.

Phase

Phase 4: ETL-First Delivery — Work Packet P4-W2

Test plan

• Verify cargo test --workspace passes (737 tests)
• Verify cargo clippy --workspace --all-targets -- -D warnings clean
• Review export job lifecycle: create → poll → download
• Confirm existing /v1/export/:wallet endpoint is unaffected
• Confirm JSONL and CSV format outputs are correct
• Verify no memory leak from String::leak() pattern

🤖 Generated with Claude Code

[user1303836]

Remote Review — P4-W2: Export Jobs

Verdict: Review pass is clear. No blocking issues found.

GHA Status

All 4 checks pass: Formatting ✓, Clippy ✓, Tests ✓, Build (release) ✓

Review Focus Assessment

1. Memory management (in-memory storage, 100k record cap) — EXPORT_MAX_RECORDS = 100_000 provides a reasonable bound. TTL pruning via prune_stale_export_jobs cleans up completed jobs. The download_export handler clones the body bytes while holding the read lock, which is necessary since the response needs owned data; acceptable given the bounded record count.

2. Content-Disposition header — Uses HeaderValue::from_str(&disposition) with proper error handling. No String::leak. Clean.

3. CSV serializers — csv_escape correctly handles commas, double-quotes, and newlines. Optional fields like token_symbol and direction use unwrap_or("") without escaping, which is safe for their typical value domains (short identifiers, enum-like values). decoded_fields (JSON map) is intentionally omitted from CSV — JSONL covers the full schema. Acceptable design choice.

4. Parquet deferred — ExportFormat enum in core is extensible. Adding a Parquet variant later won't require request model changes.

5. Shared job semaphore — Export jobs acquire from job_semaphore (10 slots shared with ingestion). This controls total concurrency but means a burst of exports could starve ingestion or vice versa. Acceptable for current scale; splitting into separate semaphores is a natural follow-up if contention appears.

6. Legacy /v1/export/:wallet preserved — Not touched in the diff. Confirmed unchanged.

Additional Observations (non-blocking)

• The 6 export query methods in v2_repo.rs delegate cleanly to existing query_* methods with EXPORT_MAX_RECORDS and offset=0. Good reuse.
• 24 new tests cover the full lifecycle: auth, validation, format selection, job status, download (completed/pending/failed), pruning, serialization, all 6 datasets, and non-exportable dataset rejection. Thorough.
• ExportFormat in core/materializer.rs includes serde roundtrip, FromStr, and Display tests.
• README and CLAUDE.md route lists are updated.

This packet is ready to merge from the PR-comment review perspective.