fix(pooling): exact within-model draw indices, keep score labels on-axis

[thomaspinder]

Summary

Two follow-ups on the pooling branch (#164), stacked on feat/151-predictive-pooling:

• Use rng.integers for within-model draw selection in _mixture_draws #165: _mixture_draws selected within-model indices via floor(rng.random(N) * sizes); now rng.integers(0, sizes[membership]) — exact (no theoretical index == size rounding edge), clearer, equally vectorised. Behaviour note: membership is drawn before the index draw, so model membership under fixed seeds is bit-identical and only within-model positions shift — all 103 pre-existing pooling tests pass unmodified, and no doc pins the old mechanism.
• plot_pool_weights score annotation can spill past the axis for weights near 1 #166: plot_pool_weights placed the score annotation at w + 1% with xlim max(1, 1.05·w), overflowing the axis for weights near 1 (reproduced: label edge ~75 px past the axis). Labels now render inside the bar (right-aligned, white) when the bar exceeds 75% of the axis, outside otherwise. New test asserts alignment and get_window_extent().x1 <= ax.bbox.x1 for both placements, red-verified against the pre-fix code.

Closes #165
Closes #166

Gates

122 pooling+plotting tests green; branch fast suite 638 passed; ruff/ty clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Egjd7ToFeb9TQqFnfRQZxV

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.7%. Comparing base (ba8129b) to head (47ef90b).

Additional details and impacted files

@@                     Coverage Diff                     @@
##           feat/151-predictive-pooling    #217   +/-   ##
===========================================================
  Coverage                         94.7%   94.7%           
===========================================================
  Files                               44      44           
  Lines                             3009    3014    +5     
  Branches                           376     377    +1     
===========================================================
+ Hits                              2851    2856    +5     
  Misses                             114     114           
  Partials                            44      44           

☔ 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.