feat: improve blob download from filestores

[mrzeszutko]

Summary

Restructures blob retrieval in HttpBlobClient to fix timing issues caused by redundant retries and improve configurability for operators.

Problem: HttpFileStore retried all HTTP errors (including 404s) with a [1, 1, 3] backoff, wasting ~5s per blob hash on "not found" responses. This compounded badly with the outer retry loop in getBlobSidecar, leading to ~5-12s retrieval times when blobs weren't yet available in filestores.

Changes:

• Configurable HttpFileStore retry/timeout (yarn-project/stdlib/src/file-store/http.ts): Added HttpFileStoreOptions with retryBackoff (empty array = no retry) and timeoutMs fields. Existing callers keep the default [1, 1, 3] retry. Blob filestore clients are created with retryBackoff: [] and a configurable timeout (default 10s via BLOB_FILE_STORE_TIMEOUT_MS), eliminating all inner retries.

• Unified retry loop in getBlobSidecar (yarn-project/blob-client/src/client/http.ts): Replaced the 3-phase flow (filestore -> consensus -> filestore-with-retry -> archive) with a single retry loop that alternates between two sources (consensus and filestore), then falls back to archive. Historical sync uses a short [1, 1] backoff; near-tip uses [1, 1, 1, 2, 2] for eventual consistency.

• Configurable source order: New BLOB_PREFER_FILESTORES env var (default: false) lets operators try filestores before consensus if their filestores are faster/more reliable.

• Supernode detection persisted (testSources() now populates superNodeHostIndexes): Non-supernode consensus hosts are skipped during blob fetching, avoiding wasted requests to hosts that can't serve blob sidecars.

• Consensus fetch no longer double-retries: fetchBlobSidecars now uses native fetch instead of the retry-wrapped this.fetch, since the outer retry loop handles transient errors.

• Documentation updated (docs/docs-operate/operators/setup/blob_storage.md): New env vars documented, retrieval flow description updated.

• Tests updated for the new fetch order, blobPreferFilestores order swap, and unified retry behavior.

Files changed

File	Change
yarn-project/stdlib/src/file-store/http.ts	HttpFileStoreOptions type, configurable retry/timeout
yarn-project/stdlib/src/file-store/factory.ts	Pass options through createReadOnlyFileStore
yarn-project/stdlib/src/file-store/index.ts	Re-export HttpFileStoreOptions
yarn-project/blob-client/src/filestore/factory.ts	Pass { retryBackoff: [], timeoutMs } for blob filestores
yarn-project/blob-client/src/client/config.ts	blobPreferFilestores, blobFileStoreTimeoutMs config
yarn-project/blob-client/src/client/http.ts	Unified retry loop, supernode filtering, tryConsensusHosts(), resolveSlotNumber()
yarn-project/blob-client/src/client/interface.ts	Simplified isHistoricalSync docs
yarn-project/blob-client/src/client/http.test.ts	Updated tests for new flow
yarn-project/foundation/src/config/env_var.ts	New env var types
docs/docs-operate/operators/setup/blob_storage.md	New env vars, updated retrieval docs

Fixes A-880

[spalladino]

Looks good! Just a small optimization.

[AztecBot]

❌ Failed to cherry-pick to v4-next due to conflicts. (🤖) View backport run.