Fix stale _transactionException after ROLLBACK TO SAVEPOINT

[kartikey321]

Fixes #454

Problem

When any PgException occurred inside runTx, the internal _transactionException field was set and never
cleared — even after a successful ROLLBACK TO SAVEPOINT restored PostgreSQL to a healthy state. mayCommit
therefore returned false and the driver sent ROLLBACK instead of COMMIT, silently discarding all subsequent
work that completed successfully at the PostgreSQL level.

Root cause

_handleMessage sets _transactionException on every ErrorResponseMessage (line 531) but has no path to clear
it. After ROLLBACK TO SAVEPOINT, PostgreSQL sends ReadyForQuery(state='T') — the authoritative signal that the
connection is in a clean, committable state — which the driver was ignoring for this purpose.

Fix

Clear _transactionException in _handleMessage when a ReadyForQuery with state='T' is received. This matches
the approach taken by asyncpg and pgx,
both of which treat ReadyForQuery('T') as the signal that the connection is healthy.

Testing

Added a regression test in test/transaction_test.dart that:

1. Triggers a 42P01 error inside runTx
2. Recovers via ROLLBACK TO SAVEPOINT / RELEASE SAVEPOINT
3. Inserts a row and expects it to be committed

The test fails on the original code and passes with this fix.

[kartikey321]

This pr changes are merged in another pr #457

Closing this one