Break WASIp3 HTTP server ownership cycle

[carsonfarmer]

Fixes #3703.

The WASIp3 handler state stored a strong reference to HttpServer. The server also owns that handler state. After the first WASIp3 request, these references kept each other alive and prevented the server and its application state from being released.

This change stores a weak reference in the handler state. Spin upgrades it while it creates a new WASIp3 worker. If the server is already gone, worker creation returns an error.

The regression test sends two single-use WASIp3 requests. It then drops the in-process server and verifies that the server was released. The test failed on upstream main before the production change and passes with this change.

Active workers can still keep the server alive until they end. This change only removes the permanent server-to-handler-to-server reference cycle.

No dependencies or services were added.

Tests

• make lint
• make build
• cargo test -p spin-trigger-http --no-fail-fast -- --nocapture
• cargo test --test integration wasi_http_p3 -- --nocapture
• cargo test --test integration wasip3_handler_does_not_retain_http_server -- --nocapture

[itowlson]

@carsonfarmer Thanks for the report and the PR. It looks good to my inexpert eye, but I've flagged a couple of knowledgeable folks for a review.

For our guidance, how urgent is this for you? And would you require it in a release, or do you pin to a SHA?

[carsonfarmer]

I'm just pinning to sha, so no rush on this honestly.

[dicej]

Thanks so much for catching, fixing, and adding a test for this, @carsonfarmer!