Unfreeze densities and replay ion density instead of density_thermal

[bclyons12]

Summary

• Add IMAS.unfreeze! calls throughout actors (fits, pedestal, core transport, EPED profiles, flux matcher) after modifying density_thermal, so that the density expression stays correct
• Replay ion density instead of density_thermal in replay actors (core transport, flux matcher, pedestal), so that total density remains constant when density_fast changes
• Add SimpleDFSane as a fallback in the flux matcher basic_polyalg and default GKNN to use basic_polyalg
• Fix STEP case to unfreeze electron density after initialization

Test plan

• Run flux matcher with TGLFNN and GKNN models
• Verify replay actors preserve total ion density when fast particle content changes
• Run STEP case end-to-end
• Run existing FUSE test suite

🤖 Generated with Claude Code

[bclyons12]

Companion to ProjectTorreyPines/IMAS.jl#292

[nanshi1177]

[bclyons12]

@nanshi1177 What shot and mode is this?

[bclyons12]

@nanshi1177 @jmcclena @tomneiser @TimSlendebroek
I reverted making :simple_dfsane the default. Here is timing, convergence, and core Te info for ActorStationaryPlasma on all of our use cases. We don't do :basic_polyalg for full TJLF since it needs gradients. Note that about half the time :simple_dfsane is faster, the other half :basic_polyalg is faster. But :basic_polyalg always converges as or more well. This makes sense because :simple_dfsane is the final fallback for :basic_polyalg now.

Green/red means better/worse than :simple_dfsane for timing and convergence, and "close to"/"far from" for core Te