Fix side-effect extraction in optVNBasedFoldConstExpr

[EgorBo]

VN-based constant propagation used to incorrectly handle side-effects - it used to pass "ignoreTrue = true" to gtWrapWithSideEffects. This could led to replacing IND(node) with a constant while ignoring the GTF_EXCEPT on the IND node itself (it's precisely what I've hit in #117583). Looks like most regressions from this change were caused by GT_ARR_LENGTH node so I made a small optimization in gtExtractSideEffList to fold unused GT_ARR_LENGTH into GT_NULLCHECK to mitigate some regressions.

[dotnet-policy-service]

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

[EgorBo]

PTAL @dotnet/jit-contrib @jakobbotsch correctness change, I've hit it with IND(obj) that was recognized as a method-table access and replaced with a constant type handle while obj was actually null, hence, no NRE were thrown.

Diffs, some improvements from the GT_ARR_LENGTH -> GT_NULLCHECK folding

[filipnavara]

Are the extra null checks in win-x86 diffs expected?

[EgorBo]

Are the extra null checks in win-x86 diffs expected?

I've realized that yeah, there are too many nullchecks in the diffs. The problem that in VN phase we have a bunch of optimizations that look through static readonly fields, RVA, etc - in some of those cases we actually know that the indirect is never faulty, but since we don't mutate the IR in the VN phase, we don't have a way to tell that to the assertprop phase. I'll put on hold this change

Even just

ROS<char> x = "hello";
var val = x[1];

now produces a nullcheck.