fix: scale beta_reg's continued-fraction bound with its parameters

[agene0001]

Stacked on #416 (whose accurate ln_gamma sets this PR's error floor); review only the last commit until that merges.

The Lentz recurrence in checked_beta_reg was capped at a fixed 140 iterations and silently returned whatever it had reached. The recurrence is slowest at the centre of the distribution, where the worst case over x grows like 5·min(a,b)^(1/3) (measured over a,b ∈ [1e2, 1e18] × 4001 values of x), so past min(a,b) ≈ 1.5e4 the answer was simply wrong. Against the exact identity I_½(a,a) = ½:

a = b	value returned	rel. error
1e5	0.49999969504	6.1e-7
1e6	0.49121972700	1.8e-2
1e7	0.21285001452	5.7e-1

Binomial::new(0.5, 2e6).cdf(1e6) returned 0.4916 against a true 0.50028 — this caps StudentsT, FisherSnedecor, and Beta cdfs too.

The bound now scales as 8·min(a,b)^(1/3) with headroom, clamped to bound worst-case work at ~10 ms. The loop still exits on convergence, so ordinary calls are unchanged (beta_reg(2.5, 2.5, 0.5): 150 ns vs 145 ns).

Plus four robustness fixes found by probing the parameter space rather than sweeping accuracy:

• an underflowed prefix short-circuits to the corresponding endpoint — exact, and avoids 0.0 × h, which is NaN whenever the recurrence overflowed (beta_reg(1e300, 1e-300, 0.5) was NaN before, on both sides of the bound change);
• x == 0 / x == 1 return 0/1 directly instead of via the symmetry test, which mapped x == 0 to 1.0 once a + b overflowed;
• the symmetry threshold (a+1)/(a+b+2) is computed scaled when a + b overflows, otherwise it collapses to 0 and sends every x down the transformed branch;
• the result stays in [0, 1] (it's a probability; a = b = 1e20 previously returned −2.56), falling back to the concentrated-limit step function if the truncated recurrence went non-finite.

Regression tests use exact identities needing no reference data — I_½(a,a) = ½ and I_x(a,b) + I_{1−x}(b,a) = 1 (both fail on the old fixed bound) — plus a 12×12×6 extreme-parameter grid asserting the result stays a finite probability.

🤖 Generated with Claude Code

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.52%. Comparing base (b152c47) to head (1c199a7).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #417      +/-   ##
==========================================
+ Coverage   95.49%   95.52%   +0.02%     
==========================================
  Files          65       65              
  Lines       15403    15457      +54     
==========================================
+ Hits        14709    14765      +56     
+ Misses        694      692       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[YeungOnion]

Closing — #456 has since landed a full rewrite of checked_beta_reg (Temme's uniform asymptotic expansion for the large-parameter prefactor, a 100k-iteration continued fraction that returns ConvergenceFailed instead of silently truncating, plus the endpoint/overflow handling this PR added). Every problem here is covered, more rigorously and with reference-validated accuracy. No remaining gap for this PR to fill.