Fix top_x_pct_share tie handling and zero-percent edge

[MaxGhenis]

Summary

top_x_pct_share used a threshold + >= slice:

threshold = self.quantile(1 - top_x_pct)
self[self >= threshold].sum() / self.sum()

This overstated the share whenever rows tied at the threshold — with constant values every call returned 1.0 regardless of top_x_pct, and top_x_pct_share(0) returned the share of the max bucket instead of 0. Downstream bottom_x_pct_share, top_10/1/0.1_pct_share, bottom_50_pct_share, and t10_b50 all inherited the bug.

Fix: sort by value ascending, cumulate weight, locate the cutoff row where the bottom 1 - top_x_pct of weight ends, and split that row's weight proportionally — the standard wealth-share algorithm.

Reproduction

import microdf as mdf

mdf.MicroSeries([5] * 10, weights=[1] * 10).top_x_pct_share(0.1)  # was 1.0, now 0.1

s = mdf.MicroSeries(list(range(1, 11)), weights=[1] * 10)
s.top_x_pct_share(0.0)  # was 10/55, now 0.0
s.top_x_pct_share(1.0)  # now 1.0

Test plan

• Existing 51 tests still pass
• New test_top_x_pct_share_handles_ties_and_edges covers:
  • Constant values (every p returns exactly p)
  • Range 1..10 (closed-form checks against p/55)
  • Unequal ties
  • bottom_x_pct_share complementation
  • Downstream top_10/top_50 still match

[MaxGhenis]

Self-review. Verified:

• _weighted_top_share algorithm: sort ascending, cumulate weights, find cutoff index via searchsorted(cum_w, target_bottom, side='right'), split the cutoff row's weight proportionally. I hand-computed the constant-series case ([5]*10, p=0.1 → target=9, k=9, top_sum=5*1=5, share=0.1) and the unequal-ties case ([1,1,10,10], p=0.5 → target=2, k=2, top_sum=10 + 10*(3-2)=20, share=20/22≈0.909). Both match expected.
• Edge cases documented and tested: p<=0 → 0.0, p>=1 → 1.0, all-zero series → np.nan. Constant-series returns exactly p for any p ∈ [0, 1].
• Algorithm is the standard wealth-share / Lorenz-curve approach; downstream bottom_x_pct_share, top_10/1/0.1_pct_share, bottom_50_pct_share, t10_b50 inherit correctly since they all delegate to top_x_pct_share.
• Docstring clarifies the tie-proportional semantics. Also fixed the bottom_x_pct_share docstring bug (said "top %" — now "bottom %").
• Tests include constant, 1..10 closed-form (top_x_pct_share(p) = floor(n*p) last rows summed proportionally), unequal ties, and downstream helpers.
• CI green Py 3.9-3.13.