Fix implicit nullchecks for floats on arm64

[EgorBo]

Fixes #65942 (comment)

[ghost]

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

Issue Details

---

Fixes #65942 (comment)

Author:	EgorBo
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[kunalspathak]

LGTM

[AntonLapounov]

@EgorBo I hit the same assertion in my PR. Is it a different issue?
https:/dotnet/runtime/pull/71661/checks?check_run_id=7203673648

  Starting:    System.Text.RegularExpressions.Tests (parallel test collections = on, max threads = 4)

Assert failure(PID 21 [0x00000015], Thread: 37 [0x0025]): Assertion failed '(shiftAmount == scale) || (shiftAmount == 0)' in 'Microsoft.CodeAnalysis.DotnetRuntime.Extensions.SyntaxValueProviderExtensions:<GetMatchingNodes>g__matchesAttributeName|4_2(byref,byref,System.String,bool,byref):bool' during 'Generate code' (IL size 286; hash 0x72e8996c; Tier1)

    File: /__w/1/s/src/coreclr/jit/emitarm64.cpp Line: 7028

[EgorBo]

@AntonLapounov could be 🤔 in theory can be caused by #71439 - it sets nullchecks to be TYP_BYTE always even if we access a local of other type - it's ok on x64 but on ARM we're can't do that, scale must match input's type. cc @jakobbotsch

[jakobbotsch]

I'll take a look tomorrow, maybe we ended up in the box removal logic with a GT_LEA child node. Though it also seems like a backend bug if it cannot handle any combination of indirection and node computing an address.