[mono][jit] Revert parts of '[mono] Remove the support for non r4fp, its not used by any supported platforms. (#82005)'

[vargaz]

Some platforms don't support r4fp, so put back the general JIT support code. Force r4fp on platforms that support it.

Fixes#84401.

[Xinlong-Wu]

Noticed you have revert this patch. I'm wondering if it is because the architecture RISC-V64 uses code like cfg->r4fp ? If yes, I will submit a patch to re-implement floating point arithmetic for RISC-V.

I had to use cfg->r4fp before because the floating point numbers in the RISC-V architecture of previous Mono would treat single floating point as double floating point during register passing, so I had to rely on cfg->r4fp to do some special processing when loading floating point immediate numbers and floating point numbers in the stack, to converting them to double floating point numbers to avoid breaks the ABI of the RISC-V architecture.

Removing the parameter cfg->r4fp causes Mono to generate the correct IR for RISC-V, which actually fixes the problem exactly. So there is no longer any need to do some extra special handling to make the RISC-V code ugly if we remove cfg->r4fp completely.

I think this is necessary, so I will submit an implementation that does not use cfg->r4fp. Feel free to close or merge it )