python: better cache in PET resolveEnv()

[austin3dickey]

Addresses part of #12500. Related to #12327, where some Python interpreters weren't getting resolved correctly after we bumped PET. Possibly will help with #12253 too, where some Python interpreters weren't being discovered in the first pass of discovery after we bumped PET.

resolveEnv() was decorated with @cache(30_000, true). But if PET hadn't discovered the env yet, undefined is pinned in the cache for 30 seconds, even after PET finished discovery. Any caller in that window got the stale undefined back. This PR replaces the decorator with an explicit cache on NativePythonEnvironments that:

1. Only caches successful PythonEnvInfo resolutions.
2. Never caches undefined or thrown errors.
3. Is invalidated in addEnv/removeEnv, so a late-arriving PET discovery automatically supersedes any prior cached state.

Release Notes

New Features

• N/A

Bug Fixes

• N/A

QA Notes

@:interpreter

Reusing the #13064 QA trick to simulate a slow PET: add

await new Promise((r) => setTimeout(r, 30000));

inside finder.refresh's loop in nativeAPI.ts.

Before this PR: picking an interpreter via the picker in the first 30s after a cold start would sometimes log Tried to switch to a language runtime that has not been registered and fail to start.

After this PR: selection should succeed as soon as PET has seen that env without a 30s penalty.

Unit tests covering the regression are included.

[github-actions]

E2E Tests 🚀
This PR will run tests tagged with: @:critical @:interpreter

^{readme  valid tags}