perf(stellar): pipeline streaming scan to overlap RPC fetch with decrypt

[collinsezedike]

Summary

• Add a generic bounded-queue pipeline(source, capacity) helper in src/chains/stellar/scanner/pipeline.ts. A background pump keeps pulling from source up to capacity items ahead of the consumer, so the next RPC fetch's I/O is already in flight while the consumer does CPU-bound work on already-buffered items, instead of alternating "await a full window, then process it."
• scanAnnouncementsStream now wraps its source in pipeline(source, window) instead of prefetching a strict window before scanning any of it. Peak memory stays O(window); cancellation and error propagation are unchanged.
• The old windowed algorithm is kept as scanAnnouncementsStreamSequential for benchmark comparisons, matching how scanAnnouncementsLegacySharedSecretTag is retained for the view-tag-batching benchmark.
• Fixed test/chains/stellar/bench/scan.bench.ts, which had corrupted/duplicated content on develop (left over from an earlier merge) making it fail to even parse — unrelated to this change but blocking it since I needed to extend that file.
• Docs in docs/chains/stellar-streaming-scan-pipeline.md, changelog entry under the upcoming release.

Closes #126.

Semver

Additive-only: new internal module (src/chains/stellar/scanner/, not part of the public package exports) plus a new benchmark-only export scanAnnouncementsStreamSequential. fetchAnnouncementsStream and scanAnnouncementsStream's public signatures and behavior are unchanged.

Test plan

• pnpm build
• pnpm test — full suite passes (793 passed in the Stellar module, no regressions)
• New test/chains/stellar/scanner/pipeline.test.ts (runs under pnpm test, not excluded like bench/): order preservation, overlap timing (pipelined run is reliably ≥25% faster than a measured sequential baseline with matched I/O/CPU delays), backpressure bound with a fast producer + slow consumer, cancellation propagation, error propagation
• test/chains/stellar/bench/scan.bench.ts — new "Stellar streaming scan pipelining" section: correctness test plus before/after benchmarks against a mock paginated RPC source; measured 36% wall-clock reduction on the 10k-announcement canned dataset (537ms → 343ms), clearing the 30% target
• Existing scanAnnouncementsStream tests (window sizing, cancellation, memory-bounded) pass unmodified
• pnpm format:check

[drips-wave]

@collinsezedike Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits