fix: MP3 flags descriptive hardware reset-state prose #834

MP3 reports a HIGH memory-manipulation finding for a descriptive hardware state:

The finite state machine reaches its reset state after two clock cycles.

Reproduced on main 4c52fbb83bbca0dd13f2685453c9b661d8c95bf8 using an actual static-only CLI scan:

skillspector scan hardware.md --no-llm --format json

Expected: this sentence describes the state of a hardware finite-state machine and should not produce MP3.

Actual: MP3 interprets the substring reset state as an agent-memory instruction.

The existing reset-state coverage exception from #453 handles a different enumeration form, and #720 handles timestamp-formatting comments. The proposed correction is narrowly scoped to descriptive hardware/FSM state grammar. Actual agent-state reset instructions and independent findings must remain detectable.

I am working on a fix with regression controls for ordinary hardware prose, direct reset instructions, and mixed content.