fix(render,network,handoff): port the litclock-dev#640/#642/#644/#645/#646 train

[kapoorankush]

Second of two port PRs for v0.224.0. Ports litclock-dev#640, #642, #644, #645, #646.

Merge order matters. This is stacked on fix/port-hotspot-password-and-clone-warning (#54) because both PRs touch CHANGELOG.md and separate branches would guarantee a conflict. Merge #54 first, and retarget this PR's base to master before you do — merging #54 with --delete-branch while this points at it will close this PR.

What changes

• litclock-dev#640 — handoff splash row clamps. Long values ran off the glass, and collided with the right column well before that.
• litclock-dev#642 — cross-file string parity enforced in CI. Strings that must match across files (handoff completion triggers, dispatcher markers) had no mechanical check, so a rename in one file silently broke a grep contract in another.
• litclock-dev#644 — QR screen clamps, and ONE derivation of the text budget. The bound that applies is the 760px text budget, not the 800px panel.
• litclock-dev#645 — record the LAN IP before the handoff gate. The dispatcher's marker write sat behind .handoff-complete, so on a fresh provision it was skipped entirely and the marker stayed absent for a whole DHCP lease, far past the 300s settling grace. Every clock coming out of first boot showed "Network: Connection issue" with a blank LAN IP.
• litclock-dev#646 — attribute the handoff completion to the path that actually did it, making the "verify all three completion paths" QA step answerable for the first time.

Port notes

• 51 bare dev#NNN refs requalified to litclock-dev#NNN.
• Two new files here: tests/test_cross_file_string_parity.py, tests/test_nm_dispatcher_behavior.py.
• One conflict, in the _anomalies.py docstring: this repo already carried the requalified litclock-dev#596 ref while the incoming commit rewrote the sentence around it. Took the newer wording (it describes what #645 actually changed) and kept the qualified ref.
• The #646 CHANGELOG hunk re-added #649's entry, which feat(wifi): persist the setup hotspot password per device; let the clone warning print #54 already ported. Dropped the duplicate; #649 appears exactly once.
• Verified nothing was dropped: of every line the five upstream commits added, the only absentees were superseded within the same series (#644 replaced #640's max_w with PANEL_TEXT_BUDGET and reformatted nine assertion f-strings). tests/test_eink_wrap.py is byte-identical to dev; src/eink_display.py differs only by this repo's existing bare-#NNN requalifications.

Testing

• ruff + shellcheck clean
• pytest: 3360 passed, 62 skipped (was 3252 on feat(wifi): persist the setup hotspot password per device; let the clone warning print #54 — the 108 new tests are the ported ones; all skips pre-existing, in untouched files)
• npm run test:js: 191 passed

Review found a pre-RC defect in the ported code — filed, not patched here

Adversarial review found that litclock-dev#645's link-local blanking disables the same-IP short-circuit: CURRENT_IP="" fails the [ -n "${CURRENT_IP:-}" ] gate that guards the exit 0, so every dhcp4-change at a 169.254.x address falls through to a full e-ink render. Measured three renders from three identical events where a real address produces one. That reinstates the unbounded-render / SPI-contention failure the #309 A4 guard was written to prevent, on a device with a finite panel-cycle budget.

It is present identically on dev, so it is fixed upstream and ported rather than patched here — patching only this repo adds to the forward divergence litclock-dev#657 already tracks. Filed as litclock-dev#667, scoped pre-RC alongside litclock-dev#660, with three smaller findings from the same review.