fix(UserMountCache): Set transaction isolation level to REPEATABLE_READ before executing insertIfNotExist without unique index

[provokateurin]

oc_mounts no longer has a unique index since #32877, so the insertIfNotExist does not work properly and depends on consistent reads always getting the same data, which is not possible with READ_COMMITTED (https://dev.mysql.com/doc/refman/9.2/en/innodb-transaction-isolation-levels.html).

The proper fix is to add back the unique index, but with a hash of the mountpoint column (which was too long for the index and was the reason the unique index was removed). This type of workaround is already used for the share_external table.

All other insertIfNotExist in server are fine, because they operate on tables with a unique index.

[SystemKeeper]

Tested it locally like this:

• Set a breakpoint at

  server/lib/private/Files/Config/UserMountCache.php

  Line 126 in 8fac473

  $this->connection->commit();

• Truncate oc_mounts
• Run occ files:mount:refresh admin with Xdebug (so it stops at the breakpoint above)
• Run occ files:mount:refresh admin without Xdebug
• Continue execution of the first occ command

Without this PR
The second command finishes before the execution of the first is continued, resulting in duplicated mounts

With this PR
The second command "hangs" and only finishes after the execution of the first one is continued. Resulting in non duplicated mounts.

[provokateurin]

This can be used as a patch to mitigate the issue, but the real fix is still under development.

[provokateurin]

Closing in favor of #56933