Improve liveness for 'STORE_BLK(lcl_var)'.

[sandreenko]

The rule for an untracked variable was weaker so converting a tracked variable to untracked sometimes gave us better code.

Don't set lvMustInit for tracked lclVars that are compiler temps if they don't contain GC fields.

Tiny SPMI improvements on master but I need it for my future changes.

tests.pmi.windows.x64.checked.mch
3 total methods with Code Size differences (3 improved, 0 regressed), 0 unchanged.
Top method improvements (bytes):
          -7 (-3.61% of base) : 247502.dasm - TailCallOptTest:Caller2(System.Object,long,double,TypedDouble,TwoInts,TypedDouble):bool
          -7 (-3.37% of base) : 247500.dasm - TailCallOptTest:Caller1(System.Object,long,double,TypedDouble,TypedDouble,TypedDouble):bool
          -7 (-3.66% of base) : 249520.dasm - Program:Caller(int,int,int,int,MyStruct):int

libraries.pmi.windows.x64.checked.mch
1 total files with Code Size differences (1 improved, 0 regressed), 0 unchanged.
          -7 (-13.21% of base) : 7338.dasm - Microsoft.FSharp.Text.StructuredPrintfImpl.LayoutOps:boundedUnfoldL(Microsoft.FSharp.Core.FSharpFunc`2[__Canon,__Canon],Microsoft.FSharp.Core.FSharpFunc`2[Nullable`1,__Canon],Microsoft.FSharp.Core.FSharpFunc`2[Nullable`1,Boolean],System.Nullable`1[Int32],int):Microsoft.FSharp.Collections.FSharpList`1[[Microsoft.FSharp.Text.StructuredPrintfImpl.Layout, FSharp.Core, Version=5.0.0.0, Culture=neutral, PublicKeyToken=b03f5f7f11d50a3a]]

Diff example:
before.txt
after.txt

[sandreenko]

@erozenfeld, could you please review it if you have time? It is not what I asked about but a different issue, the one that I asked I could not have fixed and will leave it for now.
PTAL @dotnet/jit-contrib,

[sandreenko]

I have changed the solution to better analyse "store_blk" and use "store_lcl_var" in the cases where "store_blk" was used with no reason.

It gives improvements on the same set of methods, so the original commit was dropped, maybe I will still need in the future, with additional struct changes, but for now, it does not produce diffs.

The new diffs:

benchmarks.run.windows.x64.checked.mch
1 total methods with Code Size differences (1 improved, 0 regressed), 0 unchanged.
libraries.pmi.windows.x64.checked.mch
12 total methods with Code Size differences (12 improved, 0 regressed), 0 unchanged.
tests.pmi.windows.x64.checked.mch
3 total methods with Code Size differences (3 improved, 0 regressed), 0 unchanged

[sandreenko]

Could somebody please review it @dotnet/jit-contrib ?

[briansull]

Nice improvement

[briansull]

Looks Good

[briansull]

This is good too.

[erozenfeld]

LGTM