JIT: Handle TYP_DOUBLE on arm softFP similarly to TYP_LONG

[jakobbotsch]

Avoids PUTARG_REG being a multi-reg node on arm32 to handle this special case.

The gist of the change: we expect multi-reg structs, including TYP_LONG on arm32/x86, to be put into a FIELD_LIST(PUTARG_REG(low), PUTARG_REG(high)) form before LSRA. For multi-reg structs this happens during morph, while for TYP_LONG we keep them as primitives until lowering, so it happens during lowering.

There was a special case for arm32 with softFP where TYP_DOUBLE also ends up being a similar multi-reg struct that is passed in two integer registers. There was a bunch of special casing to handle this in the backend, by making PUTARG_REG and BITCAST multireg nodes on arm32 only.
This PR instead puts TYP_DOUBLE into the expected multireg-struct form during lowering, allowing it to be handled like any other multireg struct.

Motivation is to avoid arm32-specific logic in #103866, which currently fails in CI for arm32 softFP because of this special case.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[jakobbotsch]

cc @dotnet/jit-contrib PTAL @dotnet/samsung @kunalspathak

[tomeksowi]

Thank you for the cleanup. LGTM, shouldn't affect RISC-V.

[clamp03]

Thank you.
+ Checked coreclr test on armel tizen

[kunalspathak]

LGTM. Thanks!