fix(masthead): restart_needed means a restart HELPS — behind boards get their own cue

[OrangePeachPink]

Closes #1712. All three ACs, though the diagnosis in the filing was wrong and the real one is sharper.

There is no mtime comparison

The filing guessed at "an mtime-vs-process-start comparison that trips on file touches that aren't builds." There isn't one. The flag was a single expression:

"restart_needed": bool(server.get("stale")) or bool(behind),

Observed state had server_stale: false — so behind was the trigger. That's a logical certainty from the code, not an inference: nothing else writes this flag.

Two conditions, one sentence, different remedies

condition	actual remedy
server_stale	restart the Monitor ✅
behind (board on older fw)	flash the board — restarting does nothing

The operator was told to do something that cannot work.

The part I didn't expect

The masthead already has the right cue for behind boards:

if(V.restart_needed) cue = `↻ newer build — restart the Monitor`;
else if(V.firmware.behind.length) cue = `N boards on older fw (latest X)`;

It's an else if after restart_needed. So while behind fed that flag, the accurate message was unreachable — every behind board rendered the restart prompt instead. The right words were written, shipped, and could never appear.

Fixing the flag is what lets that cue render for the first time.

The change

• restart_needed = server_stale. behind is untouched and still published in the firmware block — it just stops claiming a restart is the answer.
• The tooltip loses "or a board is behind" (the conflation, stated out loud) and now says restarting does not reflash a board.

One test asserted the bug

test_device_behind_latest_firmware_is_flagged asserted restart_needed is True for a behind board — the defect written down as an expectation. Amended rather than deleted, with the reason in place, so a reader hitting that diff sees a corrected model rather than a fixture tweak.

ACs

• AC1 — derives from the versions block's own facts. Pinned by a test asserting restart_needed tracks server_stale exactly, so a third condition can't be folded in later without bringing its own cue and remedy.
• AC2 — tree == running build → no banner. ✅
• AC3 — genuinely newer checkout still triggers it. ✅

Deliberately not changed

When the server is stale and boards are behind, the restart cue still wins the else if. That precedence predates this, and showing both is a masthead-density question rather than an honesty one — the cue that remains is true in that state. Flagging it rather than quietly widening scope.

just check green; 16 masthead tests.

Lane note: card_context.py is analytics. The board assigns this to DX and the ACs are precise, so I took it — but if Data would rather own changes in that module, say so and I'll route it instead of assuming the board settles it.

— DX