trial-params gain widen (int16) + controller-run duration (arena-fw #4)

[floesche]

Match the firmware's canonical trial_params (0x08) re-layout from reiserlab/LED-Display_G6_Firmware_Arena#4: gain widens from int8 to int16 and moves after init_pos, and a new duration field lets the controller auto-stop a trial on its own instead of relying on a host follow-up command.

• encodeTrialParams: reordered params (init_pos @5-6, gain @7-8, duration @9-10), gain now uses the existing i16le helper (int8 helper dropped), duration stays in seconds at the JS API boundary and converts to 10 ms ticks on the wire (AC::constants::duration_tick_ms, matching the firmware)
• buildTrialParams: carries duration through from the trialParams command, same convention as conditionDuration's wait/trialParams math
• hostSideTrialEnd/ArenaRunner._timer/the setTimeout mechanism is untouched — host-side timing stays authoritative for now; the wire-side duration just unblocks that swap for later, it isn't made here (kept the blast radius small until this is bench-verified)
• golden-vector tests rewritten for the new offsets and widened gain range, plus new duration vectors and range-throw cases

This is a breaking wire change — bytes 5-10 of the 0x08 payload are reinterpreted — and ships in lockstep with the companion Arena-Firmware PR. A stale pairing would silently misparse gain/init_pos/duration into garbage that still returns status == 0.

Verification

• pixi run test: full suite passes, including the rewritten trial-params golden vectors in tests/test-arena-wire-g6.js and tests/test-arena-runner-g6.js
• pixi run format-check
• Bench-verified against the paired Arena-Firmware branch: fed encodeTrialParams()'s actual output bytes to a Teensy 4.1 over serial and confirmed Mode 2 (initPos=9, gain=-20, duration=1.5s) applies init_pos correctly and auto-stops to ALL_OFF on its own, and Mode 4 gain=2000 (past the old int8 ceiling) is accepted

Test plan

• pixi run test
• pixi run format-check
• Hardware round-trip with Arena-Firmware gh4-trial-params-relayout
• Merge together with the Arena-Firmware PR, not independently

This is part of the reiserlab/LED-Display_G6_Firmware_Arena#35 PR

[mbreiser]

Resolving as superseded: this PR's change landed on main in full via #155 — commit 737f7b3 there is line-for-line identical to 1f24458 here (same patch, rebased onto v0.38 with the duty byte stacked on top). GitHub couldn't auto-mark it merged because the commit object differs, but the content is 100% on main (verified by diffing the two patches).

One heads-up on the duration half, @floesche: after bench testing on fw c3475fd we found controller-timed stops change trial behavior for protocols authored against host timing (the display blanks at duration instead of playing until the next command). So the runner currently pins the wire duration to 0 as a compat measure (ca0bec8) — the encoder still supports it, the Console bench form still sends real durations, and the plan is to re-enable it in the runner together with the run-complete timing swap you cross-referenced in arena-runner-g6.js. The gain int16 widening is fully live, including in the designer schema (±32767).

Thanks for the wire work — closing this one; #155 carries the credit.