Use a temp dir instead of mutating the CairoMakie src dir

[jkrumbiegel]

Fixes #3519

I think a temp dir is the simplest solution. These should theoretically be cleaned up by Julia before exiting, so it shouldn't litter the file system. I also think it's better to make new files rather than reuse one, because a user might look at different versions at the same time.

[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	3.41s (3.40, 3.43) 0.01+-	380.24ms (376.27, 385.06) 3.06+-	472.84ms (462.41, 512.90) 17.82+-	7.09ms (7.01, 7.14) 0.04+-	25.36ms (25.33, 25.40) 0.03+-
master	3.42s (3.39, 3.44) 0.02+-	362.47ms (304.35, 392.03) 37.95+-	465.79ms (461.58, 471.11) 3.57+-	7.09ms (7.02, 7.12) 0.04+-	25.37ms (25.33, 25.42) 0.03+-
evaluation	1.00x invariant, -0.01s (-0.41d, 0.46p, 0.01std)	0.95x invariant, 17.77ms (0.66d, 0.26p, 20.50std)	0.99x invariant, 7.04ms (0.55d, 0.34p, 10.69std)	1.00x invariant, 0.0ms (0.06d, 0.91p, 0.04std)	1.00x invariant, -0.0ms (-0.10d, 0.86p, 0.03std)
CairoMakie	3.04s (3.01, 3.09) 0.03+-	320.85ms (318.53, 328.58) 3.53+-	139.86ms (139.05, 143.49) 1.61+-	7.20ms (7.08, 7.29) 0.07+-	600.48μs (595.26, 605.85) 3.43+-
master	3.03s (3.00, 3.08) 0.04+-	320.33ms (317.78, 322.85) 2.08+-	139.93ms (138.15, 141.86) 1.32+-	7.21ms (7.13, 7.43) 0.11+-	605.67μs (601.70, 607.96) 2.05+-
evaluation	1.00x invariant, 0.01s (0.19d, 0.73p, 0.03std)	1.00x invariant, 0.52ms (0.18d, 0.74p, 2.81std)	1.00x invariant, -0.07ms (-0.05d, 0.93p, 1.47std)	1.00x invariant, -0.0ms (-0.02d, 0.97p, 0.09std)	1.01x faster ✓, -5.19μs (-1.84d, 0.01p, 2.74std)
WGLMakie	3.72s (3.65, 3.80) 0.05+-	323.89ms (316.05, 337.23) 7.10+-	8.86s (8.72, 8.99) 0.10+-	9.25ms (8.91, 9.79) 0.31+-	72.38ms (67.07, 78.33) 4.20+-
master	3.72s (3.63, 3.86) 0.07+-	319.53ms (316.35, 326.13) 3.98+-	8.90s (8.73, 9.07) 0.14+-	9.12ms (8.93, 9.38) 0.15+-	73.90ms (68.79, 81.45) 4.46+-
evaluation	1.00x invariant, 0.0s (0.01d, 0.98p, 0.06std)	0.99x invariant, 4.36ms (0.76d, 0.19p, 5.54std)	1.00x invariant, -0.04s (-0.36d, 0.51p, 0.12std)	0.99x invariant, 0.12ms (0.51d, 0.37p, 0.23std)	1.02x invariant, -1.52ms (-0.35d, 0.52p, 4.33std)

[jkrumbiegel]

@SimonDanisch any idea why this would break relocatability of glmakie?