Namespace collision between Session Replays and FileBlobs prevents S3 Lifecycle Management

There is a path collision in the underlying storage (S3) between Session Replay segments and generic FileBlobs.
- Replays use a retention-based prefix: `"{retention_days or 30}/{project_id}/{replay_id}/{segment_id}"` (e.g., 90/123/...): https://github.com/getsentry/sentry/blob/26.3.1/src/sentry/replays/lib/storage.py#L64
- FileBlobs use a hashed path: shards/path (e.g., 90/7b0d/... where 90 is a hash prefix): https://github.com/getsentry/sentry/blob/26.3.1/src/sentry/models/files/abstractfileblob.py#L218

It is impossible to configure an S3 Lifecycle Policy to expire Replays based on their retention period (e.g., prefix 90/) because this prefix also contains permanent FileBlobs (artifacts, symbols) that happen to start with the same hash.

```
$ s3cmd ls s3://sentry/sentry/90/
DIR s3://sentry/sentry/90/2/
DIR s3://sentry/sentry/90/3/
DIR s3://sentry/sentry/90/4/
DIR s3://sentry/sentry/90/7b0d/
DIR s3://sentry/sentry/90/c649/
DIR s3://sentry/sentry/90/dec7/
```

Start with the linked paths in src/sentry/replays/lib/storage.py and src/sentry/models/files/abstractfileblob.py, comparing how replay segments and FileBlobs are constructed in S3. Reproduce the overlapping prefixes with the issue's examples, then verify that replay retention can target replay objects without affecting permanent FileBlobs.

Written by the indexing model from the issue text.