feat(router): report decrypted plaintext size in ListObjects responses

[pszafarczyk]

ListObjects/ListObjectsV2 previously reported the at-rest ciphertext size (plaintext + 28-byte AES-GCM-SIV envelope). Intercept bucket-level list requests, subtract the fixed encryption overhead from every , and clamp at 0. Bucket sub-resource GETs (acl, versioning, multipart listings, ?versions, …) are still forwarded unchanged.

• cryptoutil: export EncryptionOverhead constant + invariant test
• router: bucketOnlyPattern + isListObjects, wired into getHandler
• handler: extract forwardUpstream; add handleListObjects (buffer, rewrite on 2xx, fix Content-Length)
• listsize: byte-preserving regex rewrite + unit tests
• e2e: assert v1 and v2 listings report plaintext size

Known limitation: objects not written through the proxy (legacy plaintext, server-side copies, multipart) are reported 28 bytes short / 0 after clamp, since list responses carry no per-object encryption metadata.

[louisgls]

Merged in #55 as part of the v1.9.0 release. Reviewed it end to end while integrating — the approach holds up: buffering a list page (≤1000 keys) is safe, the sub-resource exclusion list is thorough, and recomputing Content-Length after the rewrite is the detail that most implementations get wrong. Verified against MinIO through the added e2e coverage.

Three observations for a follow-up, none blocking:

1. Path-style addressing only. bucketOnlyPattern is ^/([^/?]+)/?$, which requires at least one path segment. With virtual-host-style addressing (bucket.s3.example.com/) the request path is just /, so isListObjects returns false and sizes are not rewritten. The behaviour therefore depends on how the client is configured — our own deployment uses s3forcepathstyle: true so it is unaffected, but a client using the default addressing style would silently get ciphertext sizes.

2. Content-encoded responses. If an upstream ever returns a gzip-encoded list body, io.ReadAll yields the compressed bytes, the regex matches nothing, and the response passes through with unmodified sizes. Silent no-op rather than corruption, so the failure mode is benign, but it is invisible. Checking resp.Header.Get("Content-Encoding") and skipping (or logging) would make it explicit.

3. Mixed buckets. Documented in the changelog already, just restating the operational shape: in a bucket where some objects were not written through the proxy, those are reported 28 bytes short — or 0 once clamped, for anything under 28 bytes. Fine for a bucket written exclusively through the proxy (our Loki bucket), worth a warning in the README for anyone pointing this at an existing bucket.