Follow-ups to #313: one Path Length width helper, TRANSPORT_DIRECT test, two comment fixes #322

Nits from the review of #313 (#282, merged as 30c7de46). Details are in the review comment on #313. None of them blocked the merge.

1. DRY. public/packets.js buildFieldTable, the Path Length row (Follow-ups to #268, #260 and #212: pktEsc listener leak, test gaps and comment fixes #282 point 7), has an inline second copy of senderPathHashSize()'s width rules: TRACE gives no width, 0b11 is invalid, and 0x00 on route 2/3 is the direct marker. AGENTS.md asks for one implementation. Reuse senderPathHashSize(), or extract a shared helper both call.
2. Test gap. Mutant (pkt.route_type === 2 || pkt.route_type === 3) → (pkt.route_type === 2) survives test-packets.js and test-frontend-helpers.js. Nothing covers the TRANSPORT_DIRECT 0x00 marker in the hex breakdown. Add a case. Fixing 1 and 2 together is easiest.
3. Comment. The header of test-issue-282-pktesc-listener.js and its case "a second render inside the same visit…" claim that the old closure stacked "on every filter/region change within a visit". It did not: renderLeft() returns early on those paths. Correct the wording.
4. Comment. In public/channels.js, the Follow-ups to #268, #260 and #212: pktEsc listener leak, test gaps and comment fixes #282 point 8 comment says normalizeObservedPathHashSizes() "stays" because the sender badge uses it. renderSenderPathHashBadge() does not call it. Correct the comment.