fix: release nested freezes after backup failure

[sanikolaev]

This PR prevents a failed backup from leaving an RT table frozen when an error occurs after its per-table FREEZE.

Follow-up to #138.

How to reproduce manually

rm -fr /opt/homebrew/var/manticore/t
mysql -h0 -P9306 -e "drop table if exists t; CREATE TABLE t(f text) exceptions_list='a=>b'; INSERT INTO t(id,f) VALUES (1,'a'); FLUSH RAMCHUNK t"
rm /opt/homebrew/var/manticore/t/exceptions_chunk0_0.txt
manticore-backup --backup-dir=/tmp/

 ✘  ~/manticoresearch-buddy  main  mysql -P9306 -h0 -e "show locks"
+------+------+-----------+-----------------+
| Type | Name | Lock Type | Additional Info |
+------+------+-----------+-----------------+
| rt   | t    | freeze    | Count: 1        |
+------+------+-----------+-----------------+

Root cause

Backup freezes tables twice:

1. A bulk FREEZE protects all selected tables.
2. Each table is frozen again while its files are copied.

If a frozen file disappears before it is processed, such as an exceptions file, filesize() throws before the per-table UNFREEZE is reached.

The existing shutdown cleanup releases only one freeze level, leaving the affected table locked with Count: 1.

Fix

Wrap the work following the per-table FREEZE in try/catch/finally:

• On failure, release the initial bulk freeze and rethrow the original exception.
• Always release the per-table freeze in finally.

This balances both freeze levels without querying SHOW LOCKS in production.

Test

Added a regression test that:

• Performs the real bulk and per-table freezes.
• Injects a missing file after the per-table freeze.
• Verifies that backup fails with the expected filesize(): stat failed error.
• Verifies in the same process that SHOW LOCKS contains no lock for the table.