Improve error checking for node labels

[jeremykubica]

The node strings must be unique in order to store the data internally in a GraphState. This PR changes the logic to check that each node seen during sampling has a unique string.

Also adds a few very small performance optimizations (e.g. np.array -> np.asarray).

[github-actions]

Before [18536db]	After [8255a16]	Ratio	Benchmark (Parameter)
1.54±0.1ms	1.36±0.05ms	~0.88	benchmarks.TimeSuite.time_apply_passbands
725±10μs	775±20μs	1.07	benchmarks.TimeSuite.time_fnu_to_flam
4.31±0.3ms	4.43±1ms	1.03	benchmarks.TimeSuite.time_chained_evaluate
121±1μs	124±2μs	1.02	benchmarks.TimeSuite.time_sample_x0_from_distmod
15.7±0.08μs	15.9±0.2μs	1.01	benchmarks.TimeSuite.time_sample_x1_from_hostmass
8.67±0.6ms	8.64±0.1ms	1.00	benchmarks.TimeSuite.time_evaluate_salt3_model
1.36±0.01s	1.35±0.01s	0.99	benchmarks.TimeSuite.time_make_x1_from_hostmass
28.3±0.1μs	27.7±0.4μs	0.98	benchmarks.TimeSuite.time_make_new_salt3_model
9.16±0.1ms	8.91±0.03ms	0.97	benchmarks.TimeSuite.time_load_passbands
4.88±0.1ms	4.69±0.02ms	0.96	benchmarks.TimeSuite.time_evaluate_salt3_passbands

Click here to view all benchmarks.

[OliviaLynn]

Looks good!