Migrate np.polyfit to np.polynomial.polynomial.polyfit

[danstam]

Fixes #2322

Replaces the deprecated np.polyfit API with np.polynomial.polynomial.polyfit throughout mitiq/zne/inference.py.

What changed and why

The main challenge was that the new API removed the cov=True option. The covariance matrix is now reconstructed manually using the weighted Vandermonde approach:

cov = (residuals / (n - deg - 1)) * inv((W @ V).T @ (W @ V))

One non-obvious issue: full=True silently suppresses RankWarning — numpy only raises it when full=False. A manual rank-deficiency check (rank < deg + 1) was added to preserve the existing ExtrapolationWarning behavior.

Other changes:

• Fixed coefficient ordering reversal across all affected call sites — the new API returns coefficients in ascending order [a_0, a_1, ..., a_n] vs the old descending order [a_n, ..., a_1, a_0]
• Migrated np.polyval → np.polynomial.polynomial.polyval (argument order is reversed in the new API)
• Removed [::-1] reversals in _ansatz_unknown and _ansatz_known closures
• Updated hardcoded test assertions to match ascending coefficient order
• Added cast(npt.NDArray[np.float64], ...) to satisfy mypy since numpy's type stubs for full=True return an imprecise type

All 216 ZNE tests pass. Ruff and mypy clean.

AI use disclosure

I used Claude Code to help navigate the codebase and verify the Vandermonde covariance math. All implementation decisions and code reviews were done by me iteratively.

[natestemen]

I don't want this to be a breaking change, so can you make sure the coefficients are returned in the same order as previously?

[danstam]

Added a reversal on the return value to keep the original coefficient ordering.

[danstam]

@natestemen Let me know if you need anything else.