Speed up unit conversions

[jeremykubica]

Do a single matrix multiply instead of carrying the units through the full computation.

[github-actions]

Before [88bd226]	After [600586a]	Ratio	Benchmark (Parameter)
11.0±1ms	9.57±0.5ms	~0.87	benchmarks.TimeSuite.time_evaluate_salt3_model
2.23±1ms	842±600μs	~0.38	benchmarks.TimeSuite.time_fnu_to_flam
121±2μs	125±2μs	1.04	benchmarks.TimeSuite.time_sample_x0_from_distmod
1.49±0.08ms	1.49±0.1ms	1.00	benchmarks.TimeSuite.time_apply_passbands
9.13±0.06ms	9.10±0.2ms	1.00	benchmarks.TimeSuite.time_load_passbands
15.7±0.5μs	15.7±0.2μs	1.00	benchmarks.TimeSuite.time_sample_x1_from_hostmass
4.66±1ms	4.61±0.3ms	0.99	benchmarks.TimeSuite.time_chained_evaluate
1.37±0.02s	1.36±0.01s	0.99	benchmarks.TimeSuite.time_make_x1_from_hostmass
28.8±0.5μs	28.2±0.3μs	0.98	benchmarks.TimeSuite.time_make_new_salt3_model
5.85±0.3ms	5.08±0.2ms	0.87	benchmarks.TimeSuite.time_evaluate_salt3_passbands

Click here to view all benchmarks.

[hombit]

Looks good, thank you. How much faster is this?

[jeremykubica]

Running this function in test_snia.py takes about 50% of the time (with the asarray change as well).