fix(record): record a page reload instead of dropping it as a same-URL navigation

[MaxFreedomPollard]

Problem

During recording, a page reload (F5, the reload button, or location.reload()) leaves no step in the trace. A flow that only takes effect after a refresh therefore stalls on replay, because the trace never says to reload (#139).

Root cause

observeRecordedNavigation opens with

if (!url || url === navigation.currentUrl) return { kind: "noop" };

The guard is there so the webNavigation.onCommitted / onCompleted pair of one navigation yields one step. A reload commits to the URL the tab is already on, so the same guard swallows it on the same line. The rest of the pipeline was already prepared for reloads: trace-reducer-v3.ts has carried reload: "reload" in TRANSITION_CAUSES, and the protocol defines NavigationCause::Reload, but the buffer never produced a draft that reached either.

Reproduced through the real recorder (start on https://example.com/, onCommitted with transitionType: "reload", onCompleted, stop): on main the exported v3 trace has steps: [].

Fix

One condition in step-buffer.ts: a commit whose transitionType is reload passes the same-URL guard. Only the committed event carries the transition type, so the completion that follows (same URL, no type) still collapses onto it and no duplicate is produced. The step reduces to { op: "navigate", to: <url>, cause: "reload" }, the shape the protocol already defines, so whoever replays the trace can issue bsk reload at that point.

Everything else is unchanged: action-caused navigations still annotate the action with navigatedTo, redirect hops still coalesce, and a reload triggered by pressing Enter in the address bar carries from_address_bar, which the reducer already maps to user_typed.

One consequence worth stating: Chrome reports reload for script-initiated reloads too, so a page that reloads itself is now recorded as well. Replaying bsk reload there is harmless and matches what the page actually did.

Tests

Three tests added (846 → 849), all failing on main with steps: []:

• step-buffer.test.ts: a reload of the current page is appended as a navigate draft with transitionType: "reload"; the completion that follows still collapses onto it, so there is no duplicate.
• record-steps.test.ts: end to end through handleRecordStart → onCommitted(reload) → onCompleted → handleRecordStop, the exported v3 trace has exactly one step, { op: "navigate", to: START_URL, cause: "reload" }, with pre- and post-states.

Verification

• pnpm lint — exit 0 (biome, stylelint, dsh-plugin typecheck and tests)
• pnpm --filter @browser-skill/extension compile — clean
• pnpm ext:test — 76 files, 849 passed
• pnpm ext:build — clean

Not exercised in a live Chrome here; the recorder harness in record-steps.test.ts drives the same webNavigation listeners the extension installs.

Closes #139