e2e: fail fast when every interpreter quick pick row is deprioritized

[midleman]

Summary

selectQuickInputElementContaining silently clicked the first match when every row matching the version was deprioritized. On the Linux CI images that means clicking the uv-managed base install instead of /root/.venv before it has registered -- that interpreter cannot start, so the failure surfaced 90s later as a /started/ timeout instead of at the picker. Source of a 4.3% ubuntu/electron flake in the Quarto inline-output persistence test.

• Throw when deprioritize is set and no non-deprioritized row exists, listing the rows that were seen
• The existing retry in startAndSkipMetadata now reopens the picker, giving the venv time to register
• Windows (System) and macOS (Pyenv) CI have only deprioritized rows, so they keep the first-match fallback via ALLOW_DEPRIORITIZED_PYTHON_FALLBACK
• Applied to the session-start and notebook kernel-selection call sites

QA Notes

Test-infra only; no product code. Windows and macOS lanes matter here -- they take the fallback path, not the new throw.

Validation Steps

@:quarto @:sessions @:notebooks @:positron-notebooks @:interpreter @:win @:web

E2E Triage Diagnosis

🟢 High confidence -- Test-side: when every quick-pick match is deprioritized, selectQuickInputElementContaining silently clicks the first match (the broken uv managed base interpreter), so the session never starts and the 90s /started/ wait times out.

• Test: Quarto - Inline Output: Persistence > Python - Verify kernel status persists after window reload
• Targeted failure: Pattern A -- expect(locator).toBeVisible() failed on locator('.console-instance[style*="z-index: auto"]').getByText(/started/); 6 occurrences, 4.2% on main, ubuntu/electron
• Signal: Trace timeline t=3093637 shows exactly 2 rows matching aria-label*="Python 3.10.12" (nth=0 and nth=1 read, no nth=2). Both matched DEPRIORITIZED_PYTHON_SOURCES (sessions.ts:27), so the loop at quickInput.ts:203-210 never broke out and fell through to matches.first(). Error-context snapshot, console tabpanel: "Python 3.10.12 (Global) failed to start. Could not start runtime: failed to resolve interpreter /root/.local/share/uv/python/cpython-3.10.12-linux-x86_64-gnu/bin/python". Failure line is sessions.ts:574 inside startAndSkipMetadata setup -- not in any Quarto or reload assertion.
• Frequency: 6/139 runs (4.3%) on main, ubuntu/electron
• Hypothesis: /root/.venv -- the only non-deprioritized 3.10.12 in the ubuntu24 CI image (Dockerfile.ubuntu24_04:159-176; asserted as the expected winner in remote-ssh.test.ts:85-96) -- had not yet registered as a runtime at click time. waitForInterpreterDiscoveryToComplete (quickInput.ts:94-108) only clears the "Discovering interpreters" placeholder and does not gate on the venv locator registering. The deprioritize fallback then converts "the interpreter I want is not listed yet" into "start a known-bad interpreter and wait 90s for a status that can never arrive". Ruled out: ordering flip (the image has exactly one non-deprioritized 3.10.12, so 2-rows-both-deprioritized means the venv row was missing, not reordered); never-rendered console (console-instance was present and showed explicit failure text); locator drift (/started/ is the correct locator, the session genuinely never started); Quarto/reload mechanics (failure precedes them). Surviving alternative (low): $HOME/.venv is structurally unregisterable on this image, making the third row impossible rather than late -- contradicted by the 95.8% pass rate.

[github-actions]

E2E Tests 🚀
This PR will run tests tagged with: @:critical @:quarto @:sessions @:notebooks @:positron-notebooks @:interpreter @:win @:web

Why these tags?

Tag	Source
@:critical	Always runs (required)
@:quarto	PR description
@:sessions	PR description
@:notebooks	PR description
@:positron-notebooks	PR description
@:interpreter	PR description
@:win	PR description
@:web	PR description

More on automatic tags from changed files.

^{readme  valid tags}

[jonvanausdeln]

LGTM!