apex_sim: apex-finder simulator app + sensitivity bench

[jspaezp]

New crate rust/apex_sim: an eframe/egui desktop app plus a bench for probing
the timsseek apex finder under synthetic conditions.

What it does

• Synthesizes extraction products (chromatograms) with tunable knobs:
  peak shape, per-fragment noise, theoretical-vs-observed ratio mismatch
  (obs_scale), absent/empty transitions, randomly injected peak-like objects,
  precursor isotopes, global noise floor. Deterministic (seeded ChaCha8Rng).
• Feeds them to the REAL scorer (timsseek TraceScorer / Extraction / traces).
  No scoring is reimplemented; the app runs the two production passes
  (compute_traces -> suggest_apex -> score_at).
• UI: overlaid chromatograms (colored by ion series), transitions-x-cycles
  heatmap, and all intermediate traces (cosine/scribe/lazyscore/log_int/
  ms1/apex_profile) with TRUE / pass1 / pass2 apex markers, plus a full
  ApexScore decomposition (base_score, au/cg terms, 11 features).
• Prints the final SimParams on window close for reproducibility.

Bench (examples/bench.rs, apex_sim::bench)

• Runs a catalog of canonical scenarios (clean, moderate_noise,
  high_noise+interference, heavy_interference, mismatched_library,
  absent_top_fragment) over many seeds and prints per-scenario detail plus
  a comparison summary table (pass1/pass2 sensitivity, median error, us/run).
• Simulation is built up front so it stays OUT of the scoring timer; a single
  TraceScorer is reused across runs (matches production ScoringWorker).

Fidelity

• The sim applies the same filter_zero_intensity_ions +
  select_top_n_fragments(TOP_N_FRAGMENTS) the pipeline runs before scoring,
  so it scores exactly what timsseek would. Those three items are widened to
  pub in timsseek (only visibility, no behavior change).

Build / CI

• egui/eframe/egui_plot gated behind a default-on 'gui' feature; the bench
  builds lean with --no-default-features (no egui/winit in CI).
• .github/workflows/apex_bench.yml runs the bench (--no-default-features) on
  PRs touching apex_sim/timsseek and upserts a sticky results comment.

Validation

• cargo test -p apex_sim (7 pass), and under --no-default-features (7 pass).
• cargo build -p apex_sim (gui) and --no-default-features both clean.
• Ran the bench locally; example sticky-comment output is the summary table.
• Reviewed by a second pass (senior-eng); the two fidelity/methodology
  findings (production filtering + reused-scorer timing) are applied here.

[github-actions]

Apex-finder bench

cargo run -p apex_sim --release --example bench -- 1000 2

Sensitivity + timing across canonical scenarios

=== clean ===
  config: n_cycles=245 sigma=1 noise_floor=0.02 random_peaks=200 true_apex=30
  runs=1000 tol=±2 cycles
  build :    38.32 ms total  (38.32 us/run)
  score :    29.62 ms total  (29.62 us/run, 33764 runs/s)
  scorer errors:    0
  pass1 sensitivity: 100.00%  (1000/1000)
  pass2 sensitivity: 100.00%  (1000/1000)
  pass2 |apex-truth| cycles: mean=0.00 median=0 max=0
  pass2 err histogram: 0:1000 1:0 2:0 3:0 4:0 5:0 6:0 7+:0 

=== moderate_noise ===
  config: n_cycles=245 sigma=1 noise_floor=0.2 random_peaks=50 true_apex=30
  runs=1000 tol=±2 cycles
  build :   163.51 ms total  (163.51 us/run)
  score :    28.61 ms total  (28.61 us/run, 34953 runs/s)
  scorer errors:    0
  pass1 sensitivity:  78.30%  (783/1000)
  pass2 sensitivity:  77.80%  (778/1000)
  pass2 |apex-truth| cycles: mean=21.12 median=1 max=214
  pass2 err histogram: 0:474 1:278 2:26 3:7 4:6 5:3 6:2 7+:204 

=== high_noise+interference ===
  config: n_cycles=245 sigma=1 noise_floor=0.5 random_peaks=100 true_apex=30
  runs=1000 tol=±2 cycles
  build :   292.95 ms total  (292.95 us/run)
  score :    28.46 ms total  (28.46 us/run, 35136 runs/s)
  scorer errors:    0
  pass1 sensitivity:  47.80%  (478/1000)
  pass2 sensitivity:  47.40%  (474/1000)
  pass2 |apex-truth| cycles: mean=48.45 median=7 max=214
  pass2 err histogram: 0:294 1:165 2:15 3:9 4:4 5:5 6:4 7+:504 

=== heavy_interference ===
  config: n_cycles=245 sigma=1 noise_floor=0.5 random_peaks=300 true_apex=30
  runs=1000 tol=±2 cycles
  build :   819.97 ms total  (819.97 us/run)
  score :    28.52 ms total  (28.52 us/run, 35062 runs/s)
  scorer errors:    0
  pass1 sensitivity:   8.40%  (84/1000)
  pass2 sensitivity:   8.40%  (84/1000)
  pass2 |apex-truth| cycles: mean=85.54 median=77 max=214
  pass2 err histogram: 0:53 1:28 2:3 3:4 4:11 5:8 6:8 7+:885 

=== mismatched_library ===
  config: n_cycles=245 sigma=1 noise_floor=0.5 random_peaks=100 true_apex=30
  runs=1000 tol=±2 cycles
  build :   292.88 ms total  (292.88 us/run)
  score :    28.59 ms total  (28.59 us/run, 34975 runs/s)
  scorer errors:    0
  pass1 sensitivity:  25.30%  (253/1000)
  pass2 sensitivity:  25.00%  (250/1000)
  pass2 |apex-truth| cycles: mean=71.34 median=51 max=214
  pass2 err histogram: 0:114 1:112 2:24 3:5 4:2 5:8 6:8 7+:727 

=== absent_top_fragment ===
  config: n_cycles=245 sigma=1 noise_floor=0.5 random_peaks=100 true_apex=30
  runs=1000 tol=±2 cycles
  build :   293.28 ms total  (293.28 us/run)
  score :    28.53 ms total  (28.53 us/run, 35052 runs/s)
  scorer errors:    0
  pass1 sensitivity:  10.40%  (104/1000)
  pass2 sensitivity:  10.30%  (103/1000)
  pass2 |apex-truth| cycles: mean=84.59 median=76 max=214
  pass2 err histogram: 0:24 1:65 2:14 3:8 4:8 5:8 6:8 7+:865 

=== summary (n=1000, tol=±2 cycles) ===
  scenario                    pass2%  pass1%   medErr    us/run
  clean                        100.0   100.0        0     29.62
  moderate_noise                77.8    78.3        1     28.61
  high_noise+interference       47.4    47.8        7     28.46
  heavy_interference             8.4     8.4       77     28.52
  mismatched_library            25.0    25.3       51     28.59
  absent_top_fragment           10.3    10.4       76     28.53

commit d445179