Non thermal models clean up and refactor

[samaloney]

PR Description

Less code, more readable code, and some performance improvements

• Remove some unused and unnecessary thin wrappers
• The old gauss_legendre port and the fixed_quad_batch are almost identical which should now be clearer the later is numerically better as the scaling operations are not compounded in the integration as they are done at the end - changed default to fixed_quad_batch
• Figure out source of some of the numeric constants and added comment with units
• Pull out a bunch of static constants to module level vars
• Modernised some of the code structure and eliminated unnecessary checks
• More / improved comments
• More edge case tests
• ThickTarget ~ 1.5x and ThinTarget ~2x faster

• Still importing constants from legacy should we just switch to the astropy values at this stage or also pull out the constant function from legacy

AI Assistance Disclosure

AI tools were used for:

• [ ] Code generation (e.g., when writing an implementation or fixing a bug)
• [x ] Test/benchmark generation
• [ ] Documentation (including examples)
• [x ] Research and understanding
• [ ] No AI tools were used

Regardless of AI use, the human contributor remains fully responsible for correctness, design choices, licensing compatibility, and long-term maintainability.

[samaloney]

It much easier to see that gauss_legendre and gauss_legendre do almost the same thing the only difference is when the mapping to the real limits is done before gauss_legendre or after in gauss_legendre which mean the numerical error aren't folded in to the integration

[samaloney]

Just some notes I think this code is now close the limit of the possible performance without going down the Cython or numba routes which have there own complications.

The main bottle necks are bremsstrahlung_cross_section and BrokenPowerLawElectronDistribution.density which take up ~50% of the time.

Figure shows a performance comparison between main, this branch and SSWIDL function. The main reason think_target is faster is that it doesn't require the .density method.

[KriSun95]

I like the idea of having our own specific constants (e.g., from hard-coded IDL sources and now just Astropy ones). One thought I had is that it would make sense in the future to add Astropy units to these. I realised this was useful in #293

[KriSun95]

This can be addressed in another PR but I think I included this to avoid weird values in the output. In #291 I think I remove this which forced me to try and track down when things weren't quite right.