reduce ccd/multiccd memory with nccdmax/naccdmax

[thowell]

this pr introduces nccdmax/naccdmax settings for reducing the memory requirements for ccd/multiccd. instead of allocating fields for epa and multicontact with nconmax/naconmax, allocate for the size of the ccd collider with the maximum contacts nccdmax/naccdmax.

there are potentially significant memory savings for any scene that has a ccd collider and at least one more different collider (can be a primitive collider).

implementation from #966

mjwarp-testspeed benchmarks/aloha_pot/scene.xml --nconmax=24 --njmax=128 --memory

Loading model from: benchmarks/aloha_pot/scene.xml...

Model
  nq: 24 nv: 23 nu: 14 nbody: 26 ngeom: 204
Option
  integrator: EULER
  cone: ELLIPTIC
  solver: NEWTON iterations: 100 ls_iterations: 50
  is_sparse: False
  ls_parallel: False
  broadphase: NXN broadphase_filter: PLANE|SPHERE|OBB
Data
  nworld: 8192 naconmax: 196608 njmax: 128

Rolling out 1000 steps at dt = 0.002...

Summary for 8192 parallel rollouts

Total JIT time: 0.61 s
Total simulation time: 3.27 s
Total steps per second: 2,502,983
Total realtime factor: 5,005.97 x
Total time per step: 399.52 ns
Total converged worlds: 8192 / 8192

Model memory 5.38 MiB (0.37% of used memory):
 (no field >= 1% of used memory)
Data memory 397.70 MiB (27.13% of used memory):
 geom_xpos: 19.12 MiB (1.30%)
 geom_xmat: 57.38 MiB (3.91%)
 qM: 18.00 MiB (1.23%)
 qLD: 16.53 MiB (1.13%)
 efc.J: 96.00 MiB (6.55%)
Other memory: 1062.91 MiB (72.50% of used memory)
Total memory: 1466.00 MiB (3.01% of total device memory)

mjwarp-testspeed benchmarks/aloha_pot/scene.xml --nconmax=24 --njmax=128 --nccdmax=12 --memory

Loading model from: benchmarks/aloha_pot/scene.xml...

Model
  nq: 24 nv: 23 nu: 14 nbody: 26 ngeom: 204
Option
  integrator: EULER
  cone: ELLIPTIC
  solver: NEWTON iterations: 100 ls_iterations: 50
  is_sparse: False
  ls_parallel: False
  broadphase: NXN broadphase_filter: PLANE|SPHERE|OBB
Data
  nworld: 8192 naconmax: 196608 njmax: 128

Rolling out 1000 steps at dt = 0.002...

Summary for 8192 parallel rollouts

Total JIT time: 0.58 s
Total simulation time: 3.27 s
Total steps per second: 2,505,272
Total realtime factor: 5,010.54 x
Total time per step: 399.16 ns
Total converged worlds: 8192 / 8192

Model memory 5.38 MiB (0.53% of used memory):
 (no field >= 1% of used memory)
Data memory 397.70 MiB (39.07% of used memory):
 geom_xpos: 19.12 MiB (1.88%)
 geom_xmat: 57.38 MiB (5.64%)
 qM: 18.00 MiB (1.77%)
 qLD: 16.53 MiB (1.62%)
 efc.J: 96.00 MiB (9.43%)
Other memory: 614.91 MiB (60.40% of used memory)
Total memory: 1018.00 MiB (2.09% of total device memory)

SPS: 2,502,983 -> 2,505,272
total memory: 1466.00 MiB -> 1018.00 MiB