Teardown compiler to improve rendering perf

[bluwy]

Changes

Use @astrojs/compiler teardown() API to remove the internal WASM instance and cache after we're finished with the Vite build.

Testing

All existing test should pass. !bench memory should show an improvement too, but it doesn't because the technique to capture "heap used" includes the garbage collection root (for some reason).

But when testing with docs locally, I can see the optimization working where Vite build memory gets offloaded during rendering.

Docs

n/a. internal refactor.

[changeset-bot]

🦋 Changeset detected

Latest commit: 6be32fd

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[bluwy]

!bench memory

[bluwy]

This is good for review now. The !bench memory results above should be the same as the latest commit as they're using the same compiler.

[bluwy]

Hmm. This seems to make CI tests even longer! I guess it's a double-edge sword with the teardown() API, where running builds many times causes the WASM instance to instantiate more times, slowing it down 🤔

Maybe we can skip teardown for tests for now.

[bluwy]

!bench memory

[github-actions]

PR Benchmark

Memory

	Elapsed time (s)	Memory used (MB)	Final memory (MB)
SSR build	3.68	74.28	124.60
Client build	0.13	0.17	124.78
Static generate	0.58	4.57	128.84

Main Benchmark

Memory

	Elapsed time (s)	Memory used (MB)	Final memory (MB)
SSR build	3.66	74.61	124.89
Client build	0.12	-0.20	124.78
Static generate	0.61	4.62	129.38

[natemoo-re]

Change to toggle the setting makes sense, still approved! 👍