Fix sqlite delete hang by deferring rowid deletes

[jprafael]

This attempts to fix the underlying cause for duckdb/ducklake#805.

The debugging and implementation was done by an LLM.
I don't know enough of the code base to know if this is the correct solution or if there is something else that should be done instead but it does fix the deadlock.

LLM Generated:

Summary:

• Captured a stack trace during the hang: the main thread is stuck in SQLite’s busy handler (sqliteDefaultBusyCallback → sqlite3BtreeDelete), indicating a self-deadlock between the scan and delete on the same SQLite table.
• Implemented a fix in the sqlite extension: buffer rowids during the scan, then perform deletes after the scan completes. This avoids the read cursor holding locks while deletes execute.
• The new regression test now passes in the C++ test runner.

Changes

• Updated delete implementation to buffer rowids and delete after scan completes: src/storage/sqlite_delete.cpp
• Added regression test: test/sql/storage/attach_delete_hang_repro.test

Tests

env SQLITE_TPCH_GENERATED=1 timeout 120s ./build/release/test/unittest "test/sql/storage/attach_delete_hang_repro.test"
Result: pass

env SQLITE_TPCH_GENERATED=1 timeout 120s ./build/release/test/unittest "test/sql/storage/attach_delete.test"
Result: pass

[staticlibs]

Hi, thanks for the PR! I will take a look.

[staticlibs]

Thanks!

Perhaps we don't need to keep the statement on a global state at all here, but the current version should be no worse than before.