feat(agent-card): self-declared persisted agent card, exposed on /tabs

[a-biskoazh]

What

Re-homes the mx "agent card" onto modular-main and threads it end to end. The
eleven persisted, hook-immune per-tab fields — assignment, specialty,
orchestrator, objective, current_task (bounded permalog), rounds_active,
evaluations (bounded ring), usage_count, conventions, parent_tab_id,
rehome_status — plus the RoundsActive / Eval{Tokens,Scores} / Evaluation
types, are now writable (HTTP + CLI), persisted, restored across restarts,
applied to the live runtime tab, and exposed on /tabs.

Commit 432bcc0 (fields + types on TabState) is unchanged; this PR adds the
threading on top (commit 1ad37b5).

Layers touched

Layer	Change
lib.rs	append_current_task / append_evaluation / bump_usage / parse_conventions (+ *_MAX bounds)
api.rs	CardChange enum, card_route_verb, is_rehome_state; SnapshotTab +11, TabInfo +11 (camelCase currentTaskLog/roundsActive/usageCount, snake for the rest); 4 owner-drain queues + inits; route arms (master-token only, like /context)
api/cards.rs (new)	assignment / parent / rehome / card_verb / evaluation / bump_usage setters
headless.rs	card fields on HeadlessTab; restore from TabState; persist back (the ..default() would otherwise wipe them); snapshot builder; drain of the 4 queues onto the runtime tab
app.rs (GUI parity)	same: struct fields, from_state restore, both persist builders, snapshot builder, drain
api/tabs.rs	populate the card fields on /tabs (currentTaskLog = current_task)
cli/*	set-assignment, set-rehome-status + 8 card verbs (set-specialty/-orchestrator/-objective/-current-task/-rounds-active/-conventions/-evaluation, bump-usage) via a shared tab_field runner; wired into mod/client/dispatch
assets/openapi.yaml	the 11 card routes declared

built == wired (the central trap)

The card fields are not merely persisted — they are hydrated and applied:

• restore reads them from TabState onto the runtime tab (headless + GUI);
• persist writes them back (explicit, not ..default());
• the main-loop drain applies the four queues (pending_assignment/parent/ rehome/card_changes) onto the runtime tab, mirroring the existing
  context/meta drain.

Proven by round-trip integration tests (real TCP server via spawn_server):
POST a card route → 200, the value is mirrored into the snapshot, queued
for the drain, and comes back out on /tabs. Covered: assignment
round-trip, current-task append + currentTaskLog surfacing, bump-usage
increment + usageCount surfacing, rehome 400-on-bad-state.

Verification (all green)

• cargo build headless,energy ✓ · headless,energy,catbus ✓
• cargo test headless,energy,catbus → 501 passed, 0 failed
• cargo clippy (nursery=deny) headless,energy ✓ · headless,catbus ✓ · gui,energy,catbus ✓ (GUI parity proven — no allow added)
• cargo fmt --check ✓
• openapi.yaml parses, 11 card paths present

Deliberate deviations (flagged for review)

1. role_of + parse_assignment are foundation helpers published ahead of
   their consumer (commit 82d16e1). Downstream branches (task/aligator/catalog)
   each carried a private copy and drop theirs to consume this one after rebase
   (coordination MAS/tichef). Both carry a documented #[allow(dead_code)] — the
   same foundation-allow idiom upstream already uses for dashboard_url_for_role
   / rehome_badge; the allow comes off when the first in-tree consumer lands.
2. is_rehome_state inlined the 4 slugs instead of porting mx's GUI-only
   RehomeStep/REHOME_STEPS struct (labels/badges are dashboard concerns).
   Same validation, no dead struct.
3. Pre-existing, out-of-scope: cargo clippy --features gui,…,pets trips a
   suboptimal_flops nursery lint in src/pet.rs:1022 — byte-identical at the
   base commit 432bcc0, under the unrelated pets feature. Not touched.
   The GUI branches of app.rs are validated via --features gui,energy,catbus.

Not done here (follow-up candidates)

• The parent_tab_id stamping from dispatch --new (the delegate) is not wired
  to auto-POST /parent; the route + field + CLI exist, the auto-stamp on spawn
  is a separate change.

[codecov]

Codecov Report

❌ Patch coverage is 48.64479% with 360 lines in your changes missing coverage. Please review.
✅ Project coverage is 52.38%. Comparing base (239178d) to head (82d16e1).

Files with missing lines	Patch %	Lines
src/cli/set_card.rs	0.00%	85 Missing ⚠️
src/api/cards.rs	45.63%	81 Missing ⚠️
src/app.rs	12.50%	77 Missing ⚠️
src/cli/tab_field.rs	55.97%	59 Missing ⚠️
src/lib.rs	56.41%	17 Missing ⚠️
src/cli/set_rehome.rs	45.83%	13 Missing ⚠️
src/cli/client.rs	0.00%	10 Missing ⚠️
src/cli/dispatch.rs	0.00%	10 Missing ⚠️
src/api.rs	97.27%	4 Missing ⚠️
src/cli/set_assignment.rs	0.00%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #43      +/-   ##
==========================================
- Coverage   52.41%   52.38%   -0.03%     
==========================================
  Files          92       97       +5     
  Lines       28268    28969     +701     
==========================================
+ Hits        14816    15176     +360     
- Misses      13452    13793     +341     

Flag	Coverage Δ
rust	52.38% <48.64%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.