Converge the default continuous inverse_cdf instead of a fixed 16 iterations

[gaoflow]

The ContinuousCDF::inverse_cdf default runs a fixed 16-iteration bisection with no
convergence test, so it only ever narrows the initial bracket by 2^16 regardless of how
wide the bracket is or how small the quantile is. Every continuous distribution without a
closed-form quantile inherits this — currently Chi and InverseGamma — and the result
sits far from the crate's own DEFAULT_RELATIVE_ACC (1e-14):

• Chi::new(1).inverse_cdf(1e-12) returns 3.05e-5 instead of 1.25e-12; in fact every
  p <= ~1e-6 returns the same 3.05e-5 bracket floor. Even the median is only accurate
  to ~1e-5.
• InverseGamma shows the same ~1e-5 baseline error, growing into the tails.

This is the continuous mirror of the discrete side, which already converges via
internal::integral_bisection_search.

The fix replaces the fixed count with a bisection that runs until the bracket agrees to
DEFAULT_RELATIVE_ACC (capped at 100 iterations and by float resolution). Two smaller
changes make the tails accurate:

• Bracket from the distribution's own min()/max() when finite, only doubling out from
  ±2 when a bound is infinite. Chi/InverseGamma are supported on [0, ∞), so the old
  [-2, 2] seed wasted the lower half.
• In the upper half, invert sf rather than cdf. As cdf saturates to one it can no
  longer place the quantile (many x map to a single f64 cdf value); sf stays well
  conditioned. Distributions that don't override sf fall back to 1 - cdf, so this is a
  no-op for them.

Across p ∈ [1e-12, 1-1e-12] the worst relative error for Chi and InverseGamma drops
from ~1e7 (and ~1e-5 at the median) to <= 1.3e-14. Reference quantiles in the new tests
come from scipy, cross-checked against mpmath at 60 digits; a round-trip test covers a
wider grid.

let q = Chi::new(1).unwrap().inverse_cdf(1e-12); // was 3.05e-5, want 1.25e-12

[codecov]

Codecov Report

❌ Patch coverage is 99.05660% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 94.94%. Comparing base (7f6c9ee) to head (49983c2).
⚠️ Report is 10 commits behind head on main.

Files with missing lines	Patch %	Lines
src/distribution/mod.rs	97.95%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #390      +/-   ##
==========================================
+ Coverage   94.73%   94.94%   +0.21%     
==========================================
  Files          59       59              
  Lines       13052    13146      +94     
==========================================
+ Hits        12365    12482     +117     
+ Misses        687      664      -23     

☔ 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]

Thanks!

would you be willing follow this with something similar to the brent-like approach that uses NR steps that you introduced for Gamma in #382?