(bug) Legacy Railway preview setup fails on re-run: 'A volume is already mounted' treated as throttling

[mmabrouk]

When the Railway preview setup job re-runs against an already-bootstrapped preview project, bootstrap fails even though the environment is healthy. The Postgres volume add returns:

A volume is already mounted on service Postgres in environment production
Volume add at /var/lib/postgresql/data for service 'Postgres' failed (Railway often throttles volume creation); retrying in 15s (attempt 1/3).
...
Failed to create volume at /var/lib/postgresql/data for service 'Postgres' after 3 attempts.

Expected: an already-mounted volume is the desired end state, so ensure_volume should treat this response as success (or its railway volume list pre-check should detect the existing volume; today it misses it and the refusal is misclassified as the volume-creation throttle).

Reproduce: run the preview chain for a PR once (green), then use "Re-run all jobs" on the same run. First runs on fresh projects are unaffected, which is why this only shows up on re-runs. Observed on runs 30767675762 and 30767675393 (2026-08-03).

Workaround: close and reopen the PR so cleanup deletes the project and bootstrap starts fresh.

Note: the clone-based preview redesign (#5650) retires this code path entirely; this issue matters only while the legacy path remains the default.

[linear-code]

AGE-4061

[mmabrouk]

Fix implemented in the #5650 stack (bottom PR #5658): ensure_volume now treats Railway's "already mounted" refusal as success instead of misreading it as the creation throttle, so setup re-runs against existing previews converge. Verified by the stack's own preview chains once merged.