F-41: Seesaw encoder read timing loses rotation #44

Severity: LOW · Effort: Small · Batch: 5 · Status: Code done in #21 (29b54d2), awaiting bench verification

Found in the second full review of the firmware. Recorded with the rest of the pass (F-19…F-44) in docs/REVIEW_REMEDIATION_PLAN_2.md, which is where its status is kept.

The double read compensates for a missing write → delay → read sequence (the Seesaw needs time
to prepare a response). Rotation during the 5 ms between reads is lost. Split each transaction
into a write, a short delay and a read, and drop the double read — bench-verify before merging.

Acceptance Criteria

• Every Seesaw read is a write, a 500 µs wait (Adafruit's library uses 250 µs), and a
  read, each its own transaction. Every register is read once.
• Bench, before merging: slow single clicks each register exactly one step, a fast
  spin tracks without lost steps, and the button still presses and releases. If reads
  fail or return garbage, the F-19 bound drops them, so the knob stays still rather than
  jumping.

Related: #9 (moving the display port off the legacy I2C driver). The encoder HAL uses the same legacy bus, so the two should be done with each other in mind.