perf: use pseudo-Pade approximation for the spergel HLR computation

[beckermr]

This PR adds a pseudo-Pade approximation for the Spergel HLR computation. This code runs about 100x faster than the root finding code, sometimes even faster than that. Here is an example:

# nu = 2.7
root finding time: 4.613792 ms
pade time: 0.052833 ms

The notebook dev/notebooks/spergel_hlr_flux_radius_approx.ipynb has the code that fits the pseudo-Pade approximation.

[codspeed-hq]

CodSpeed Performance Report

Merging #130 will improve performances by ×120

_{Comparing spergel-hlr-pade (2dead7f) with main (c96d517)}

Summary

⚡ 2 improvements
✅ 19 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	spergel-hlr-pade	Change
⚡	test_benchmark_spergel_init[compile]	1,901.2 ms	16 ms	×120
⚡	test_benchmark_spergel_init[run]	1,527.3 µs	81.6 µs	×19