feat(kinetics): too-few-points warning, intercept checks, 'Pseudo-first (initial drop)' model

[Hj1308]

Summary

• Tab 1 warns when a catalyst has fewer than MIN_FIT_POINTS informative points (t = 0 anchor not counted): AICc then over-penalises 2-parameter models and cannot evaluate 3-parameter ones. The summary gains an n (fitted) column.
• Tab 2 checks linearisation intercepts against theory (95 % CI, anchor excluded) via the new catlab.metrics.linear_intercept_check. A PFO line that misses the origin warns about a fast initial removal step and estimates its size (100·(1 − e^(−a))). A PSO mismatch is noted only when its straight line fits (R² ≥ 0.95).
• New model "Pseudo-first (initial drop)": C(t) = A·C₀·e^(−kt), 0 ≤ A ≤ 1 — a fast initial step followed by first-order decay; A = 1 is plain PFO and is flagged as at-bound. Reports k, A, initial drop %, r₀ and t½.

Review fixes on top of the original commits

• Golden master: the new A/A_se keys were compared at rtol = 1e-9 and A_se differs by 3.6e-9 between Windows and Linux; they join the fragile parameter keys. Existing golden values are unchanged (verified programmatically); only the new model's entries were added and no golden dataset's selected model changed.
• Tab 2: the Zero-order intercept note fired on every decelerating run and repeated the PFO warning, so that check was dropped; the PSO note now requires R² ≥ 0.95.

Verification

• On 5-point data with a fast initial step, the warnings fire exactly where expected (too-few-points everywhere; intercept warning only on the runs with an initial drop).
• Synthetic two-stage data (8 points): the new model is selected in ≥ 45/50 seeds; on pure PFO data in ≤ 5/50.
• Tests use only synthetic datasets (tests/synthetic_data.py); no experimental data in the tree or the branch history.
• 188 tests pass; coverage 84 % (floor 81 %); ruff check / ruff format --check clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GN1ANT89MKDjHRXcnjgSjo

---

Generated by Claude Code