Fix the pre-existing astropy/pandas/shapely benchmark workflows

[atharvas]

Summary

Repairs the three pre-existing, perpetually-red benchmark workflows (Test Astropy/Pandas/Shapely Benchmarks).

Three distinct root causes were found and fixed:

1. Workflows auto-disabled. All three were disabled_inactivity (GitHub disables scheduled workflows after 60 days idle) — which is also why they never ran on PR SQLite snapshot storage, docs site, and release workflow #6. Re-enabled via gh workflow enable; this PR triggers them.

2. py3.8 collection bug. tests/test_cli_roundtrip.py used list[Path] annotations without from __future__ import annotations, raising TypeError: 'type' object is not subscriptable at collection time on Python 3.8 — erroring out the py3.8 leg of all three workflows (and violating the repo's own stated convention). Added the future import.

3. equal_nan=False flagged deterministic NaNs as changes. The dominant real failure class (e.g. stats.sigma_clipping.*) was Arrays not close: max_diff=nan: benchmarks that deterministically return NaN-containing arrays "failed" because the comparator defaulted to equal_nan=False. For snapshot testing, a NaN that reappears unchanged is not a regression. Changed ComparisonConfig / SnapshotConfig / CLI fallbacks to default equal_nan=True (equal_nan=False still available for strict numpy.isclose semantics). Behavior change → version bumped 0.2.0 → 0.2.1.

4. Roundtrip gate is now regression-based, not perfection-based. Real third-party suites contain inherently un-snapshotable benchmarks (memory addresses in reprs, timing-sensitive, dtype-unstable) that can never pass a "zero failures" assertion. test_cli_roundtrip.py now runs list → capture → baseline → verify and asserts the baseline→verify transition matrix has no pass-to-fail / skip-to-fail (read from verify --summary). Consistently-broken benchmarks stay fail-to-fail and are tolerated; genuine regressions are still caught. This uses the tool's own baseline subcommand + transition matrix.

Local validation

Suite	Before	After
shapely roundtrip (both tests)	py3.8 collection error	PASS (82s)
astropy core shard	127 / 114 pass / 8 fail	PASS — 0 regressions (13.5 min)
fast unit suite	—	155 passed, 8 skipped
ruff check + ruff format --check	—	clean
comparator tests (incl. new NaN coverage)	—	60 passed

Test plan

• CI: tests (py3.8–3.13) + ruff
• CI: Test Shapely/Astropy/Pandas Benchmarks (re-enabled) go green on this PR