fix(db): a failed reconnect leaks a pool lease, and slots are never released #1306

Problem

In the three pools, acquire increments the entry's refs and then awaits the reconnect of a dropped connection:

entry.refs++;
if (!entry.connection.isConnected) await _reconnect(entry, mode);
return PgLease._(this, entry, entry.connection);

• lib/core/database/postgres_connection_pool.dart:101 and :134
• lib/core/database/mysql_connection_pool.dart:89 and :122
• lib/core/database/sqlite_connection_pool.dart:93 and :127

When the reconnect throws (server restarted, network down), no lease is returned, so nobody releases it and refs stays raised. The entry is never idle-disposed (refs == 0 is required) and never evicted by _evictIfNeededBeforeNewSlot. Repeated failures across databases / modes fill the 32 slots with unevictable entries, and every later acquire fails with StateError: ... pool exhausted.

Scope

• Decrement refs (and re-arm the idle timer) when the reconnect fails, in all three pools; or take the lease only after the connection is usable.
• Test with a fake connection whose connect fails: after N failed acquires the pool can still evict and create entries.

Acceptance

• A failed reconnect leaves refs as before the call.
• After a failed reconnect the entry is idle-disposed after the delay.
• maxEntries failed acquires do not exhaust the pool.