fix(setup): camera wizard UX follow-ups from #262 review (#263)

[Adityakk9031]

Closes #263.

Addresses three non-blocking camera UX follow-ups deferred from the #262 review:

1. Duplicate camera detection: Maps candidate and assigned paths to physical camera identity (_CameraNode.camera / sysfs USB device directory or canonical resolved node) before comparing, catching duplicates across by-id and by-path aliases.
2. Fresh listing row rescan after replug: Rescans inventory and rebuilds by_id_devices and by_path_devices after a successful u unplug-to-identify flow, clearing stale rows.
3. Toggle advertisement cleanup: Suppresses the p prompt when by_id_rows == by_path_rows.

Verification

• ruff check . — All checks passed.
• ruff format --check . — All 123 files formatted.
• pytest tests/test_setup.py — 113 passed, 43 skipped (gated on POSIX/symlinks).

[Adityakk9031]

@aris-zhu and @jeqcho have a look

[jeqcho]

Thanks @Adityakk9031 — this lands all three follow-ups from #263 cleanly, and the physical-identity duplicate check in particular is exactly the shape the issue hoped for: mapping both sides through _CameraNode.camera with a resolved-path fallback is robust, and the by-id-vs-by-path duplicate test genuinely exercises it (it fails on main, as a good regression test should). The in-place clear()/extend() rebuild after u is also the right call, since the row lists are shared by reference across slots.

Two things I'd like addressed before merge, plus a couple of optional notes:

1. test_camera_view_state_suppresses_toggle_when_rows_are_identical doesn't exercise the new code (tests/test_setup.py, ~line 3819). Its inventory record has by_id set and a unique serial, so the pre-existing inventory branch already returns advertise_path_toggle=False — the test passes without this PR's change. To cover the new suppression line, use a record with by_id=None (and a by-path name): the old code advertises the toggle for it, both views then render the identical by-path row, and the new by_id_rows == by_path_rows guard is what flips it to False.

2. Reprint the active listing when the post-u rescan changes the rows (src/inspect_robots/_setup.py, ~lines 567-575). The numbered listing is printed once at section start; after a replug re-rolls udev links, the rebuilt rows can diverge from what's on screen, so a number typed at a later slot silently selects a different device than the one displayed — the very race this change is meant to close. A cheap guard: capture the old active rows, and if they differ after the rebuild, call _print_camera_listing for the active view.

Optional, non-blocking:

• A small test for the rescan path (scripted u with a rescan_inventory that returns changed rows, asserting the row lists were rebuilt) would protect item 2 of the issue, which currently has no coverage.
• advertise_path_toggle isn't recomputed after the rescan, so a replug that changes by-id availability leaves the p advertisement stale for later slots — fine to defer, just noting it.
• Tiny style nit: physical_id is re-defined on every loop iteration and the kind == "v4l2" and inventory guard appears twice; hoisting the helper above the loop would tidy this.

Also a heads-up that the branch conflicts with main (the _duplicated_serials → _ambiguous_identities refactor touched _camera_view_state), so the rebase will need a little care in exactly the function this PR edits.

Thanks again — with the test fixed and the listing reprint added, this is good to go.