fix: restore LOW AUTHORITY PNG label lost in PR #151 reformatting (+ deduplication)

[nerdCopter]

Regression

Commit `afffd36` added an orange `[LOW AUTHORITY] max=Xdps — recommendations unreliable` label to the PNG legend (base P:D section, before Conservative/Moderate/Aggressive entries) and a matching console warning in the Optimal P section.

PR #151 label reformatting removed the PNG label from the base P:D section. The console Optimal P warning still exists but is buried inside the `--estimate-optimal-p` block with no effect on recommendation output.

Fix

Restore `[LOW AUTHORITY]` label-override in both PNG and console, and limit base P:D recommendations to Conservative-only when LOW AUTHORITY fires.

Base P:D recommendations (PNG + console)

• Restore orange `[LOW AUTHORITY] max=Xdps — recommendations unreliable` label before the recommendation entries (as in `afffd36`)
• Conservative: retain — a small incremental adjustment is directionally safe even on noisy data; append `[LOW AUTHORITY — verify before applying]`
• Moderate / Aggressive: suppress — replace with `Recommendation (Moderate): Skipped [LOW AUTHORITY]` and `Recommendation (Aggressive): Skipped [LOW AUTHORITY]`

Optimal P section (PNG + console)

• Apply the same `[LOW AUTHORITY]` label consistently. Currently the console prints a bare inline warning and the PNG shows nothing; both should carry the label.

Deduplication

`afffd36` recomputed `is_low_authority` independently in `plot_step_response.rs` and `main.rs`. Compute it once after the step-response loop, cache in `[bool; AXIS_COUNT]`, and share across both sections:

// After step-response loop, before base P:D section:
let mut low_authority: [bool; AXIS_COUNT] = [false; AXIS_COUNT];
let mut low_authority_max_sp: [f32; AXIS_COUNT] = [0.0; AXIS_COUNT];
for axis_index in 0..ROLL_PITCH_AXIS_COUNT {
    if let Some((_, _, max_sps)) = &step_response_calculation_results[axis_index] {
        let max_sp = max_sps.iter().cloned().fold(0.0_f32, f32::max);
        low_authority_max_sp[axis_index] = max_sp;
        low_authority[axis_index] = max_sp < LOW_AUTHORITY_SETPOINT_THRESHOLD_DEG_S;
    }
}

Related

• `afffd36` — original implementation reference (PNG label in base P:D section)
• PR feat: Optimal P source provenance and unified console/PNG label format #151 — where the regression was introduced
• perf/feat: eliminate double-parse and make optimal-P estimation default (no flag required) #146 — double-parse elimination; unified pipeline would make LOW AUTHORITY a natural per-axis output for all downstream logic