sanity checks for texture atlas hashing & defaults

[SimonDanisch]

closes #3264

[MakieBot]

Compile Times benchmark

Note, that these numbers may fluctuate on the CI servers, so take them with a grain of salt. All benchmark results are based on the mean time and negative percent mean faster than the base branch. Note, that GLMakie + WGLMakie run on an emulated GPU, so the runtime benchmark is much slower. Results are from running:

using_time = @ctime using Backend
# Compile time
create_time = @ctime fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @ctime Makie.colorbuffer(display(fig))
# Runtime
create_time = @benchmark fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @benchmark Makie.colorbuffer(fig)

	using	create	display	create	display
GLMakie	3.56s (3.53, 3.60) 0.02+-	415.47ms (403.52, 429.85) 8.03+-	491.89ms (481.11, 508.45) 10.93+-	8.00ms (7.83, 8.07) 0.09+-	26.12ms (26.01, 26.27) 0.09+-
master	3.58s (3.54, 3.61) 0.02+-	562.94ms (556.07, 573.57) 6.02+-	710.10ms (689.80, 752.06) 20.44+-	8.00ms (7.74, 8.23) 0.15+-	26.10ms (25.91, 26.19) 0.10+-
evaluation	1.00x invariant, -0.02s (-0.75d, 0.19p, 0.02std)	1.35x faster✅, -147.47ms (-20.77d, 0.00p, 7.03std)	1.44x faster✅, -218.21ms (-13.31d, 0.00p, 15.69std)	1.00x invariant, 0.0ms (0.01d, 0.98p, 0.12std)	1.00x invariant, 0.02ms (0.25d, 0.65p, 0.09std)
CairoMakie	3.00s (2.98, 3.02) 0.02+-	331.40ms (327.52, 336.71) 3.21+-	143.49ms (141.03, 147.53) 2.29+-	7.52ms (7.43, 7.67) 0.08+-	602.91μs (596.85, 606.25) 2.99+-
master	3.02s (2.97, 3.05) 0.03+-	335.43ms (327.64, 350.68) 7.49+-	149.91ms (146.22, 155.23) 3.49+-	7.60ms (7.48, 7.69) 0.07+-	606.38μs (604.10, 613.34) 3.25+-
evaluation	1.00x invariant, -0.01s (-0.67d, 0.24p, 0.02std)	1.01x invariant, -4.03ms (-0.70d, 0.23p, 5.35std)	1.04x faster ✓, -6.41ms (-2.17d, 0.00p, 2.89std)	1.01x invariant, -0.07ms (-0.96d, 0.10p, 0.07std)	1.01x invariant, -3.47μs (-1.11d, 0.06p, 3.12std)
WGLMakie	3.65s (3.60, 3.73) 0.05+-	548.94ms (538.39, 559.64) 7.48+-	8.79s (8.64, 8.95) 0.12+-	9.72ms (9.44, 10.17) 0.26+-	72.59ms (67.36, 81.33) 5.37+-
master	3.68s (3.60, 3.84) 0.09+-	402.89ms (329.10, 590.25) 113.41+-	8.85s (8.67, 9.22) 0.18+-	10.12ms (9.48, 13.27) 1.39+-	68.62ms (67.37, 72.11) 1.62+-
evaluation	1.01x invariant, -0.03s (-0.48d, 0.39p, 0.07std)	0.73x slower❌, 146.05ms (1.82d, 0.01p, 60.45std)	1.01x invariant, -0.05s (-0.34d, 0.54p, 0.15std)	1.04x invariant, -0.4ms (-0.40d, 0.48p, 0.83std)	0.95x noisy🤷‍♀️, 3.97ms (1.00d, 0.10p, 3.49std)

[SimonDanisch]

Great that we tested this, since we do have cache misses because of JuliaArrays/StaticArrays.jl#1222.
Solved for now by writing out the bezierpath values instead of calculating them in each julia version on first call to default_marker_map.