Add inference test and improve inference of _SpectralElementGrid2D

[charleskawczynski]

This PR changes inference failures from 351 -> 189 in _SpectralElementGrid2D.

The main problem here, I think, was that the compiler was boxing CoordType2D, so this PR moves that logic to a separate function (sometimes the compiler struggles with understanding types if a variable is computed in multiple places, even in the context of union-splitting).

[charleskawczynski]

Some benchmarks, main:

julia> @benchmark Grids._SpectralElementGrid2D(Spaces.topology(space), Spaces.quadrature_style(space); enable_bubble=false)
BenchmarkTools.Trial: 206 samples with 1 evaluation.
 Range (min … max):  22.108 ms … 186.010 ms  ┊ GC (min … max): 0.00% … 87.63%
 Time  (median):     22.855 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   24.227 ms ±  11.390 ms  ┊ GC (mean ± σ):  3.85% ±  6.40%

   ▁▆▇█▁                     ▃
  ▃█████▅▄▄▆▄▄▃▃▃▃▁▃▃▁▃▁▃▃▄▄▆█▃▄▇▄▄▃▆▃▃▄▃▃▁▁▁▃▃▁▁▁▃▁▁▁▁▁▁▃▃▃▃▃ ▃
  22.1 ms         Histogram: frequency by time         26.9 ms <

 Memory estimate: 4.71 MiB, allocs estimate: 185908.

This PR

julia> @benchmark Grids._SpectralElementGrid2D(Spaces.topology(space), Spaces.quadrature_style(space); enable_bubble=false)
BenchmarkTools.Trial: 331 samples with 1 evaluation.
 Range (min … max):  13.639 ms … 175.915 ms  ┊ GC (min … max): 0.00% … 91.53%
 Time  (median):     14.057 ms               ┊ GC (median):    0.00%
 Time  (mean ± σ):   15.071 ms ±   8.921 ms  ┊ GC (mean ± σ):  4.22% ±  6.06%

     ▆█▄
  ▄▄████▅▄▆▅▃▃▄▂▃▄▂▁▁▁▁▁▂▂▄▅▄▃▃▃▄▂▄▃▃▃▃▃▃▂▁▂▁▂▃▁▁▁▁▁▁▂▃▁▁▁▂▁▁▂ ▃
  13.6 ms         Histogram: frequency by time         17.9 ms <

 Memory estimate: 3.83 MiB, allocs estimate: 135198.