[Yuren0312]

Fixes #111.

Problem

bottleneck and wasserstein only filter non-finite death times, never birth times. A diagram containing a point with a -inf birth (which can arise, e.g. from ripser's lower_star_img when the image holds -inf anchor pixels, or from a negated superlevel persistence) therefore leaks -inf into the cross-similarity matrix:

• In bottleneck, -inf - (-inf) evaluates to NaN, which lands in D and becomes ds[-1], so the returned distance is silently NaN.
• In wasserstein, pairwise_distances produces NaN for the -inf coordinate, and scipy.optimize.linear_sum_assignment raises ValueError.

Fix

Filter on both coordinates — np.isfinite(S[:, 0:2]).all(axis=1) — so a non-finite birth is handled the same way a non-finite death already is: the point is dropped with a UserWarning, consistent with the existing convention. The warning text is updated to "non-finite birth or death times" (the change is reflected in the four existing test_inf_deathtime matches).

Tests

• Updated the warning matches in the existing test_inf_deathtime tests (bottleneck + wasserstein).
• Added test_inf_birthtime to both TestBottleneck and TestWasserstein, covering a [-inf, 0.5] point and asserting a finite result (the exact 0.5 bottleneck distance, and a finite symmetric wasserstein distance).

Verification

• pytest (full suite) → 114 passed
• ruff check persim/ and ruff format --check persim/ → clean