Numpy 2+ support

[BBC-Esq]

This PR is relatively simple. As it turns out, the codebase was essentially compatible with Numpy 2+ already. For a fuller analysis, see my outline here:

#925

There was one "recommended" (but not mandatory) change, however, that will improve improves numerical stability that I implemented. Under the new "type-promotion" rules (NEP 50) says that when a Python float (default float64) is combined with a float32 array, the result’s dtype follows the higher-precision scalar—i.e., the array is up-cast to float64.
By writing

thr = np.float32(1e-5)
mask = np.abs(wav) > thr          # both operands now float32

Basically, we keep the comparison in float32, preserving:

• Memory footprint – no temporary 64-bit copy is created.
• Speed – SIMD paths for float32 aren’t lost.
• Numerical parity – results match the older 1.26 behavior.

Again, the tweak isn't mandatory, but it keeps the code's code’s performance and dtype expectations identical after upgrading to NumPy 2 +.

Please change the requirements to numpy any version LESS THAN 3...and then when 3 is released we can revisit the issue of compatibility. Thanks.

[BBC-Esq]

I'm not sure why this workflow would have failed but let me know your thoughts.

[fumiama]

I will check it. It seems not your fault.

[BBC-Esq]

Thanks, when will a new release be done so that "pip check" doesn't show a conflict?

[fumiama]

You can upgrade and check it now.