perf/feat: eliminate double-parse and make optimal-P estimation default (no flag required)

[nerdCopter]

Context

Two coupled improvements deferred from #145:

1. Double-parse (performance)

When --estimate-optimal-p is used, each log file is currently parsed twice:

1. Phase 1 (profile_aircraft_group()): parses all logs in the aircraft group to extract throttle-punch ratios for the torque-inertia profile.
2. Phase 2 (process_file()): parses the same file again for step-response analysis and optimal-P estimation.

This is a performance issue, not a correctness bug.

2. Remove --estimate-optimal-p flag (UX)

The flag was added conservatively while the feature was experimental. The analysis requires only setpoint[3] (throttle channel), which is present in nearly all Betaflight/EmuFlight logs. Gating on setpoint[3] availability rather than an explicit flag would make the feature automatic when data is present.

The feature label Optimal P (Experimental, log-derived) already signals experimental status to users.

Why these are coupled

Removing the flag means the profiler always runs when step-response output is requested. This makes the double-parse fix essential — without it, every log would be parsed twice unconditionally, which is unacceptable as a default behaviour.

The natural resolution is: unify the two parsing passes so each log is read once, then remove the flag and gate on setpoint[3] availability.

References

• Introduced in feat: Physics-based optimal P estimation with prop-size-aware recommendations #132; carried into feat: torque-inertia profiler & optimal P estimation (squash-rebase onto #143) #145
• Noted as known limitations in the feat: torque-inertia profiler & optimal P estimation (squash-rebase onto #143) #145 PR description