feat(chunk-upload): Assemble DIFs exclusively to Objectstore

[lcian]

Follow-up PR, and stacked on #120183.

This updates the chunk-upload/assemble flow to stop relying on FileBlob/File when organizations:objectstore-debugfiles-exclusive-write, paving the way for a complete migration of DIFs.
Legacy (non-chunked) DIF upload methods have been already addressed in #120183.

Under this new flag, chunk-upload works as follows:

1. On the first call to the assemble endpoint, we perform HEAD requests to Objectstore for all the chunks that belong to debug files to determine existence.
   To prevent race conditions where the file might be missing at the finalization step, files that expire in <= 1 hour are still reported as missing.
   For all the missing chunks, we record an intent in Redis chunk-upload-objectstore:{org-id}:{hash}, which allows us to tell which chunks need to go to Objectstore when they're uploaded via chunk-upload.
2. When a chunk is uploaded to the chunk-upload endpoint, if the intent is present in Redis, we additionally write that chunk to Objectstore at /chunk-upload/org={org.id}/{hash} with a 25 hours TTL (-1 hour for the check above = 1 day).
   Note that the chunks are still being uploaded as FileBlobs, as they might be needed for other kinds of assembled files that are not DIFs, which currently don't support Objectstore as backend.
   The Redis intents serve to prevent the inverse from being true: only if a chunk is needed for DIFs then it will be uploaded to Objectstore.
3. In the assemble task, we spool the chunks from Objectstore to a temporay file on disk, then use that file validation with Symbolic and subsequent upload to Objectstore as the final materialized DIF, creating a ProjectDebugFile row with just the storage_path.

A new constant MAX_OBJECTSTORE_DEBUG_FILE_SIZE is introduced, initially matching the existing MAX_DEBUG_FILE_SIZE but ready to be increased or made into a feature flag to support larger debug files.

Note: an additional Redis key chunk-upload-objectstore-mode:{organization_id} with values 0/1 is introduced in an effort to make this a graceful rollout without failures for customers.
The first DIF manifest writes this key with a 1 day TTL. If the feature flag is off, it pins that org to legacy FileBlob assembly.
If we don't do this, then it's possible that the flag gets enabled from an org between the first manifest request and the final assembly request. The assemble task would then fail to find chunks in Objectstore, and fail with an unrecoverable error.
After the flag is fully rolled out for 1 day, we can remove this mechanism and we need to make sure we clear the assemble status, so that the user can immediately retry the upload of a chunk that disappeared from Objectstore due to TTL (which should not happen normally, see point 1 above).

Close FS-365

[linear-code]

FS-365

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit e5d288f. Configure here.}

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 0c470f5. Configure here.}

[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.}

^{Reviewed by Cursor Bugbot for commit 9dfd79f. Configure here.}

[lcian]

Superseded by #121065, where we continue writing FileBlobs and assemble to Objectstore.
Even though the approach in this PR makes it easier to extend to other types of files, it only increases cost for no other benefit.