perf: faster kval for Lanczos interpolated images via cached splines

[beckermr]

This PR implements faster kval computations for Lanczos kernels via a spline interpolant.

[codspeed-hq[bot]]

CodSpeed Performance Report

Merging #101 will degrade performances by 21.38%

_{Comparing faster-interp (be95a0a) with main (ec2153e)}

Summary

⚡ 6 improvements ❌ 1 regressions ✅ 4 untouched benchmarks

:warning: Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	main	faster-interp	Change
⚡	test_benchmarks_lanczos_interp[kval-conserve_dc-compile]	5,028.7 ms	129.5 ms	×39
⚡	test_benchmarks_lanczos_interp[kval-conserve_dc-run]	42,142 µs	340.6 µs	×120
⚡	test_benchmarks_lanczos_interp[kval-no_conserve_dc-compile]	491.5 ms	129.4 ms	×3.8
⚡	test_benchmarks_lanczos_interp[kval-no_conserve_dc-run]	4,143.6 µs	443.8 µs	×9.3
⚡	test_benchmarks_lanczos_interp[xval-conserve_dc-run]	975 µs	778.3 µs	+25.28%
❌	test_benchmarks_lanczos_interp[xval-no_conserve_dc-run]	731.1 µs	929.9 µs	-21.38%
⚡	test_benchmarks_metacal[run]	1,345.3 ms	212.4 ms	×6.3