Refactor emitInsSizeSVCalcDisp to more closely match the emitOutputSV checks

[tannergooding]

This is a follow up from #117288 where I noticed a lot of complexity in how emitInsSizeSV handled it's logic and some variance between it and emitOutputSV.

There are a decent number of places where we have smaller size predictions and so allocate less bytes for the method. There are likewise a miniscule number of places where we actually emit smaller instructions based on the predictions:

Windows x64

Overall (-760 bytes)
MinOpts (-740 bytes)
FullOpts (-20 bytes)

Linux x64

Overall (-23 bytes)
MinOpts (+0 bytes)
FullOpts (-23 bytes)

However, there is a significant throughput improvement across the board due to the reduced complexity:

Windows x64

Overall (-0.18% to -0.01%)
MinOpts (-0.43% to -0.15%)
FullOpts (-0.03% to -0.01%)

Linux x64

Overall (-0.01% to +0.00%)
MinOpts (-0.08% to +0.00%)

Windows x86

Overall (-0.08% to -0.01%)
MinOpts (-0.23% to -0.10%)
FullOpts (-0.02% to -0.01%)

[dotnet-policy-service]

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

[Copilot]

Pull Request Overview

This PR refactors emitInsSizeSVCalcDisp to unify prefix and displacement size calculations and align them with emitOutputSV logic.

• Consolidates separate prefix (emitGetAdjustedSize/REX) and displacement sizing into a single pass
• Simplifies local vs. parameter frame‐offset computation
• Integrates EVEX and VEX displacement compression handling

Comments suppressed due to low confidence (2)

src/coreclr/jit/emitxarch.cpp:5154

• Add unit tests covering both successful and failed EVEX compression paths (i.e., when TryEvexCompressDisp8Byte returns true and false) to ensure dspInByte and overall returned sizes are correct.

            if (TryEvexCompressDisp8Byte(id, dsp, &compressedDsp, &dspInByte))

src/coreclr/jit/emitxarch.cpp:5113

• The original handling for temporary locals (when var < 0) has been removed; you need to reintroduce that branch or handle temps explicitly to avoid breaking frame offset calculations for stack temps.

    int  adr;

[tannergooding]

CC. @dotnet/jit-contrib. This is ready for review, it is both a small simplification to the xarch emitter and an improvement (across the board) to both code size and throughput.