Fix FTS dumps and cross-thread pool reuse

[supermomonga]

Summary

• exclude Turso internal FTS indexes from SQL structure dumps
• add an ngram FTS dump/load regression test
• verify Japanese substring search after reloading the dumped schema
• reconnect pooled Active Record connections when they are checked out by a different thread or fiber
• add a pool-size-one regression test for sequential cross-thread connection reuse

Context

The turso Ruby driver binds a native connection to the thread and fiber that created it. Active Record may safely check a pooled adapter back in and later lend that same adapter to another execution context. Reopening only the native connection during the checkout callback preserves Active Record's pooling contract without sharing a native connection across threads or fibers.

Tests

• bundle exec rake test (115 runs, 205 assertions)
• bundle exec rake test:conformance (37 runs, 63 assertions)

[bendangelo]

Thank you for your contribution.

[supermomonga]

@bendangelo
Thanks! btw, sorry I forgot to make this PR from draft to ready 😅

[bendangelo]

No problem. I'm trying to test this out in a rails app but I've run into multiple issues, like the sqlite file staying locked. and that's preventing me from continuing work. I'm open to any other contributions.

[supermomonga]

oops, are those issues caused by this PR?

Yes, I'm interested in contributing to this gem and turso gem.
When those gems are production ready, I want to switch from SQLite + Litestream to Turso Cloud in my Rails apps.

[bendangelo]

No those were not caused by you. The issue is with the turso rust code, it's something I cannot touch. I'll keep testing this on my end.