Server-side retry for conflicting rename operations

Follow-up from #4646.

When two rename operations happen at the same time, Polaris currently returns HTTP 503 (TARGET_ENTITY_CONCURRENTLY_MODIFIED). While this works, 503 usually means the server is unavailable and may cause clients or proxies to apply unnecessary backoff.

A better approach would be to retry conflicting rename operations on the server side so that temporary conflicts can be resolved automatically instead of returning an error to the client.

References:

* PR: https://github.com/apache/polaris/pull/4646
* Dev discussion: https://lists.apache.org/thread/tr8zh8121t2jb41s0q2yd9s73y2tp2tq

Review PR #4646 and the linked development discussion to locate the rename conflict handling that returns TARGET_ENTITY_CONCURRENTLY_MODIFIED. Trace how concurrent renames are detected and determine the retry behavior needed; done means temporary conflicts are retried server-side instead of returning HTTP 503 to the client.

Written by the indexing model from the issue text.