Handle read-only remounts (erofs) disk errors better

On Linux when disks / filesystems fail they can be remounted as read-only. Disk writes through Erlang's file IO in that case returns an `{errors, erofs}` error. CouchDB doesn't handle a worker returning that error well, and returns 500 responses even if there is a quorum of other workers which are able to satisfy the request.

In cases when quorum logic applies, we should be able to tolerate a worker returning `{erorr, erofs}` and perhaps treat it as a `rexi_EXIT maintenance_mode` error.

A few places to handle it:
https://github.com/apache/couchdb/blob/main/src/fabric/src/fabric_doc_update.erl#L61-L114
https://github.com/apache/couchdb/blob/main/src/fabric/src/fabric_rpc.erl#L155

Start with the quorum-handling paths in src/fabric/src/fabric_doc_update.erl (lines 61-114) and src/fabric/src/fabric_rpc.erl (around line 155), then trace how worker errors are classified. Confirm the expected behavior for Erlang read-only filesystem errors and ensure quorum requests tolerate them when other workers can satisfy the request instead of returning HTTP 500.

Written by the indexing model from the issue text.