squeeze some more space out of VODE's memory footprint

[zingale]

No description provided.

[zingale]

microphysics tests: http://groot.astro.sunysb.edu/Microphysics/test-suite/gfortran/2025-10-22-001/index.html

[zingale]

this actually results in a 1% slowdown on Frontier (for flame_wave)

[zingale]

groot GPU tests: http://groot.astro.sunysb.edu/Microphysics/test-suite/gpu/2025-10-25-001/index.html

it seems like this can make a small diff in some cases with CUDA.

On Frontier, it is less clear.

[zingale]

changing the type of the pivot array makes things worse. If I remove that, the remainder of the changes are mostly just making things const.

[zingale]

this still causes answer diffs on Frontier. Not sure where the issue is