The proptest of the gamma function fails spuriously in CI

[JSorngard]

In #41 it was discovered that the test CI job fails randomly in CI. Specifically, it fails in the proptest of the gamma function.
In that PR it fails when ran on the Rust beta channel, but I have also seen it fail on stable on my own machine.

[Axect]

The core issue seems related to the nature of proptest itself. Because proptest generates a wide range of random inputs, it's likely hitting edge cases where minor floating-point differences between our gamma implementation and C's tgamma exceed the current relative tolerance (rel_eps = 1e-10). This randomness explains why the failures appear intermittently.

As a potential fix, would you mind trying to increase the tolerance slightly? Adjusting rel_eps to 1e-9 in test_gamma_proptest might provide enough margin to account for these small discrepancies found by the random inputs.

If you could apply this change and see if it stabilizes the CI tests, that would be very helpful in confirming if tolerance is the primary issue here.

Separately, I really appreciate all the effort you're putting into making the CI robust – thank you! Thanks also for looking into this specific issue.