[stable34] fix(metadata): bind chunked file ids in dropMetadataForFiles

[dispather]

Manual backport of #62331 to stable34.

MetadataRequestService::dropMetadataForFiles() splits $fileIds into chunks of 1000 and iterates over them, but the IN clause still binds the full $fileIds array instead of the current $chunk. Each iteration therefore issues the same full-size DELETE, which produces the More than 1000 expressions in a list are not allowed on Oracle. entry from QueryBuilder::prepareForExecute() and repeats an identical statement once per chunk.

The sibling method IndexRequestService::dropIndexForFiles() carried the same copy/paste bug and was already backported to stable34 in #62771. This applies the equivalent one-line fix to the metadata table.

Difference from the master patch

master's #62331 switches the chunk size to IQueryBuilder::MAX_IN_PARAMETERS, which does not exist on stable34. This PR keeps the existing literal 1000 and changes only the bound variable, matching the shape of the already-merged stable34 backport #62771 (+1/-1, same file family).

getMetadataFromFileIds() a few lines above also binds an unchunked $fileIds, but master's #62331 leaves it untouched as well, so it is out of scope here.

Why this is filed by hand

/backport to stable34 was requested on #62331 twice without response — in prose on 2026-08-08 and as the bot command on 2026-09-06. Both requesters have author_association: NONE, so the command does not appear to be honoured from non-members. Filing manually rather than leaving stable34 unpatched; happy to close this if backportbot is triggered by a maintainer instead.

Refs #62331, #62771, #62325