chore: post-merge hardening, coverage, and 2.1.0

[dsfulf]

Follow-up to #20 (SE gamma fix) and #21 (n_grid), landing changes that weren't part of either contributor PR.

Hardening

• n_grid guard — _integrate_with now raises ValueError for n_grid < 2 (a degenerate grid), validated before the empty-t early return.
• Docstrings — n_grid is documented on the public cum() and the sibling volume methods (interval_vol, monthly_vol, monthly_vol_equiv), which previously claimed kwargs were "currently unused, reserved for future use." monthly_vol_equiv now threads **kwargs through its prepend call, matching interval_vol.

Tests

• Cover the n_grid < 2 guard and the SE small-n gamma-overflow fallback path (the parametrized SE regression only exercised n ∈ [0.3, 0.8], i.e. the closed-form branch).

Error analysis

• docs/numerical_integration.rst + integration_validation.py gain an n_grid convergence characterization (second-order; 10,000 → ~2e-6, 2,000 → ~5e-5 worst-case relative error), a note on the SE small-n fallback, and a print_ngrid_convergence() helper so the table is reproducible. Accuracy is governed by n_grid, not the density of the requested t array.

Coverage

• Upload coverage to Coveralls via coverallsapp/github-action@v2 (ubuntu/3.12 cell). pytest now writes coverage.lcov. The upload step is continue-on-error because the test matrix is a required check on main — a Coveralls outage must not block merges. coverage/coveralls can be re-added as a required check once it reports.

Version

• Bump 2.0.0 → 2.1.0. The SE fix is flagged in the version history as a breaking numerical change (EUR/cum for any SE fit with n ≠ 0.5 now differ — understated up to ~64% at n=0.3, overstated ~10% for n > 0.5).

Verified locally: ruff clean, mypy clean, 39 tests pass.