perf(objectstore): Pass through compressed bytes when client accepts encoding

[jan-auer]

Upgrades to objectstore-client 0.1.0, which adds an accept_encoding parameter to Session.get(). When the stored encoding matches a value the client accepts, the SDK skips decompression and reports the encoding in response metadata.

Both objectstore-backed endpoints now take advantage of this to avoid unnecessary decompress-then-recompress round-trips.

Proxy endpoint (organization.py)

• Parses the Accept-Encoding header and passes it to Session.get()
• When the response carries a Content-Encoding, checks whether the browser accepts it before forwarding; decompresses and strips the header if not
• Strips Content-Length when decompressing, since the value refers to the compressed size

Attachment download endpoint (event_attachment_details.py / eventattachment.py)

• Adds BlobStream dataclass and EventAttachment.get_blob_stream() to the model — for pure V2 blobs this negotiates encoding with the objectstore; all other blob types (inline, V1, V1+V2 double-write) delegate to getfile() with no encoding
• download() is reduced to a single path: call get_blob_stream(), stream the payload, set Content-Encoding or Content-Length based on whether encoding was negotiated
• Removes the double-write read verification scaffolding and the old two-path structure

Shared

• Adds parse_accept_encoding() to the objectstore module — parses the Accept-Encoding header per RFC 7231, normalises to lowercase, strips q-values, and excludes q=0 encodings

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[lcian]

Looks good to me.
I tested this and the optimization doesn't work when going through the hybridcloud proxy (because that one always uses decode_content=True for the response part), however the header is always consistent with the actual returned data format, so correctness-wise it's fine (and we shouldn't use the control silo endpoint anyways generally).