fix: realistic Cameron, altitude, splits and heat models

[0jonjo]

Phase F1a of the physiology review. Four models returned unrealistic numbers; this PR fixes them. Method names, signatures and return shapes don't change, and neither does the structure of environmental_factors.yml. Only the returned values move, except that Cameron now rejects distances above 100 km. No version bump: all phases ship together in one later release, so the CHANGELOG entry sits under [Unreleased].

Changes

1. Cameron (cameron_predictor.rb): the old a + b·e^(−d/c) constants were not Cameron's formula. They gave 10K 42:00 → marathon 2:57:34, which is more optimistic than Riegel (3:13:12). This PR replaces them with Dave Cameron's velocity-ratio model, f(d) = 13.49681 − 0.000030363·d + 835.7114 / d^0.7905, T2 = T1·(D2/D1)·f(D1)/f(D2), with d in metres. Sources: Cameron's own metric version posted to the t-and-f list (2001-06-20, https://www.mail-archive.com/t-and-f@lists.uoregon.edu/msg11312.html) and the had2know.org Cameron calculator, which states metres and gives a worked example (3.5 mi in 51:30 → 5 mi in ~75:08). That example is pinned in a test. CAMERON_A/B/C are replaced by named constants for the new formula. Domain guard: f(d) crosses zero near 445 km, so a new constant CAMERON_MAX_DISTANCE_KM = 100.0 makes every Cameron method raise ArgumentError when either distance is above 100 km. The model is fitted up to the marathon, and 100 km keeps '100k' usable.
2. Altitude (YAML): threshold_meters 914.4 → 300, plus a new 300: 0.0 point, so the penalty ramps up instead of jumping at 915 m. The NCAA points are kept exactly. New extrapolated points 3000: 7.92, 3500: 9.97, 4000: 12.2 come from the quadratic fit p = 0.3647·x² + 1.9482·x, x = km − 0.3, and the YAML documents them as extrapolation. The 0: 0.0 point is removed.
3. Splits: negative/positive pacing changes from ±4% to ±1% per half. A 3:00:00 marathon now splits 1:30:54 + 1:29:06.
4. Heat: new points 35: 8.7 and 40: 10.9 continue the 25→30 °C slope of 0.44 points/°C and are marked as extrapolation. duration_factor and the ideal range are untouched.

Compatibility checks

• No public method removed or renamed, and no required parameter added. duration_factor is unchanged, and so are the race_splits signature and strategy names.
• YAML keys are unchanged (altitude.threshold_meters, altitude.data_points, heat.ideal_range_celsius, heat.data_points). Points were only added, apart from the redundant 0: 0.0. A test pins this structure.
• Breaking (CHANGELOG › Breaking): the public constants CameronPredictor::CAMERON_A/B/C are removed, and Cameron raises above 100 km.

Baseline diff (gem_realismo/baseline.rb, main → this branch)

Only the intended sections changed. Age grade, VO2max/zones, Riegel, VDOT, even splits and heat ≤ 30 °C are identical.

Predictions, Cameron column only (Riegel | Cameron | VDOT):
5k 20:00 -> 10k:       42:26   -> 41:39
5k 20:00 -> half:      1:29:42 -> 1:31:51
5k 20:00 -> marathon:  2:59:25 -> 3:15:11   (Riegel 3:11:49)
5k 30:00 -> marathon:  4:29:08 -> 4:52:47   (Riegel 4:47:43)
10k 42:00 -> half:     1:28:47 -> 1:32:36
10k 42:00 -> marathon: 2:57:34 -> 3:16:46   (Riegel 3:13:12)
10k 60:00 -> marathon: 4:13:40 -> 4:41:06   (Riegel 4:36:00)
half 1:45 -> marathon: 3:30:00 -> 3:43:07   (Riegel 3:38:55)
(other Cameron rows move by under 1 min, e.g. 10k 42:00 -> 5k 19:47 -> 20:09)

Altitude: 500 m 0.0 -> 0.46 | 914 m 0.0 -> 1.41 | 2800 m 5.9 -> 7.2 | 3600 m 5.9 -> 10.42
          (0, 915, 1200, 1500, 2000, 2438 m unchanged)

Heat (20/45/60/120/180/240/300 min):
35 C: 3.25/4.88/6.5/13.0/19.5/29.25/29.25 -> 4.35/6.53/8.7/17.4/26.1/39.15/39.15
40 C: 3.25/4.88/6.5/13.0/19.5/29.25/29.25 -> 5.45/8.18/10.9/21.8/32.7/49.05/49.05

Splits:
negative: 1:33:36 | 26:00  -> 1:30:54 | 25:15
positive: 1:26:24 | 24:00  -> 1:29:06 | 24:45

Verification

• bundle exec rake test: 534 runs, 0 failures.
• bundle exec rubocop: 44 files, no offenses.
• The README and docstring examples are recomputed. test_documented_examples checks the single-line ones.

Draft: do not merge before review. Merging to main publishes to RubyGems.

[0jonjo]

Superseded by #97 (released as 2.0.0).