S3 on Forge

[frrist]

Track the prototype-to-production path for an S3-compatible API on top of the Forge stack. The S3 listener (codename ms3t) currently lives at pkg/ms3t/ in this repo. See pkg/ms3t/architectural.md for the prototype-level design notes, the choice points, and open questions.

Why sprue, not a separate repo? A pragmatic prototype-stage call: keep ms3t in-process with sprue so it inherits sprue's identities + clients with zero auth coordination. Up for re-evaluation once we settle a tenant model.

In scope for this epic

• Define a sensible S3 API subset for the spike #37 — Define a sensible S3 API subset for the initial spike
• Bootstrap S3 facade: HTTP listener + MST + Forge integration #38 — Bootstrap S3 facade (HTTP listener + MST + Forge integration) — anchored on feat: add embedded ms3t S3 listener backed by Forge #34
• Bucket mutation data storage: investigate separate space for block diffs #39 — Bucket mutation data storage: investigate "separate space for block diffs"
• Add ms3t to smelt + S3 smoke tests #40 — Add ms3t to smelt + S3 smoke tests
• ms3t telemetry + Grafana dashboards #41 — ms3t telemetry + Grafana dashboards
• Spike: staging deployment + production infra preparation #42 — Stand up staging deployment + production infra prep (spike)
• Spike: tenant management API #43 — Flesh out tenant management API (spike)

Status snapshot

#	Title	Status
#37	Define S3 API subset for the spike	proposed
#38	Bootstrap S3 facade	in flight (#34)
#39	Bucket mutation data storage proposal	proposed
#40	ms3t in smelt + S3 smoke tests	proposed
#41	ms3t telemetry + Grafana dashboards	partial (#35, storacha/smelt#23)
#42	Staging deployment + production infra (spike)	proposed
#43	Tenant management API (spike)	proposed

Explicitly deferred

• Sharding. Assume all data received is <256 MB. The sharding proposal in flight is fil-one/RFC#2; revisit once it lands. Adjacent prior art: storacha/RFC#65 (Filepack archive format), storacha/RFC#66 (Virtual DAG in Sharded DAG Index), and shard.rfc in this repo.
• Multipart upload, aws-chunked body decoding, AWS Sig V4 auth, pre-signed URLs, server-side encryption, x-amz-meta-* headers, versioning, object lock — these are FilOne hard requirements but are out of scope until Spike: tenant management API #43's spike proposes a sequencing.
• GC of unreferenced bodies, multi-instance / sticky-bucket routing, direct-passthrough to piri — see pkg/ms3t/architectural.md#future-directions.