Toggle pane shortcuts and upgrade Wrangler

[max-sixty]

Repeating g T, g A, or g L now closes the pane it opened, matching each pane's visible toggle. The active go-to menu names the close action, and browser regressions cover Threads, Asks, and All leaves, including an answered Asks tray.

The PR also upgrades Wrangler from 4.129.0 to 4.130.0, including its matching Miniflare and workerd versions. This follows the third recorded intermittent wrangler dev exit in the site verification gate, the escalation threshold documented in #503.

Validated with the full navigation file, the everyday Python suite, Worker typecheck and tests, dry-run deployment, pre-commit, current-base browser comparison, and the bundled-site Chrome probe. The first CI attempt hit that known Wrangler exit; its unchanged rerun passed before the upgrade.

This was written by Codex on behalf of max-sixty

[leaf-agent]

The change reads correct and I found nothing to raise on it. tests/test_render_navigation.py passes whole on this head (114 passed), and the layer's contracts hold: every consumer of does and line already reads them through word(), so making the three panel rows state-dependent changes nothing downstream; descriptorFor evaluates the return frame before run, so an opening press still names the surface it enters rather than the close wording; and a closing press leaves active() false, so invoke skips the push and prune() pops the frame the opening press left. Each close matches its own visible control exactly — showTray(openTray(key) ? null : key) for the trays, if (panelIsOpen()) setPanel(false) for Threads — which is the parity the PR set out to get.

What withholds the approval is that test is red on this commit (run 34439522486), not anything in the diff. verify_cross_tab_activation reported /examples/triage-board/ never presented, with one 500 and then thirty consecutive ERR_CONNECTION_REFUSED on 127.0.0.1:8787; the wrangler log ends on its own top-level handler with an empty ✘ [ERROR]. That is the local wrangler dev process exiting mid-walk, not a page failing to start — the same page presented in 969 ms during the profile phase of this same run, seconds earlier. main is green at 7723ed1 and every other pull-request ci run in this window passed. I have re-run the failed job.

Worth a look past this PR: this is the signature #503 diagnosed, and #503 named this exact condition for escalating — "A third — particularly another exit rather than a stall — is where wrangler dev itself becomes the thing to change, by pinning or upgrading it". This is that third, on a separate day from #475 and #503, and an exit rather than a stall. The run was again on wrangler 4.129.0, with 4.130.0 available.

[leaf-agent]

The rerun passed on the identical commit — test is green on 6629511, which settles the failure above as the wrangler dev exit rather than anything in the diff. Nothing in my review withholds the change; I can't convert the review itself to an approval without a new commit to anchor it to, so read the assessment above as clean.