JIT: remove inlining restriction for some methods that throw

[AndyAyersMS]

We were blocking inlines for methods that throw with more than one thing on the
evaluation stack. There was already logic for the non-inlining case to flush
the stack and preserve pending side effects. So we can simply remove the
inlining restriction.

Fixes #2156.

[AndyAyersMS]

cc @dotnet/jit-contrib

PMI diffs shows a few cases where inlining is unblocked; all cause code size increases.

PMI CodeSize Diffs for System.Private.CoreLib.dll, framework assemblies for x64 default jit
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: 566 (0.00% of base)
    diff is a regression.
Top file regressions (bytes):
         180 : System.Composition.AttributedModel.dasm (29.80% of base)
         161 : System.Private.CoreLib.dasm (0.00% of base)
         133 : System.Security.Cryptography.Pkcs.dasm (0.03% of base)
          63 : System.DirectoryServices.Protocols.dasm (0.07% of base)
          29 : System.Runtime.Numerics.dasm (0.04% of base)
5 total files with Code Size differences (0 improved, 5 regressed), 160 unchanged.
Top method regressions (bytes):
         161 (22.39% of base) : System.Private.CoreLib.dasm - CustomAttributeData:get_NamedArguments():IList`1:this
         133 (36.34% of base) : System.Security.Cryptography.Pkcs.dasm - PkcsHelpers:DeepCopy(CmsRecipientCollection):CmsRecipientCollection
          96 (457.14% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor(String):this
          84 (400.00% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor():this
          63 (14.58% of base) : System.DirectoryServices.Protocols.dasm - AddRequest:.ctor(String,String):this
          29 ( 7.67% of base) : System.Runtime.Numerics.dasm - BigNumber:HexNumberToBigInteger(byref,byref):bool
Top method regressions (percentages):
          96 (457.14% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor(String):this
          84 (400.00% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor():this
         133 (36.34% of base) : System.Security.Cryptography.Pkcs.dasm - PkcsHelpers:DeepCopy(CmsRecipientCollection):CmsRecipientCollection
         161 (22.39% of base) : System.Private.CoreLib.dasm - CustomAttributeData:get_NamedArguments():IList`1:this
          63 (14.58% of base) : System.DirectoryServices.Protocols.dasm - AddRequest:.ctor(String,String):this
          29 ( 7.67% of base) : System.Runtime.Numerics.dasm - BigNumber:HexNumberToBigInteger(byref,byref):bool
6 total methods with Code Size differences (0 improved, 6 regressed), 244490 unchanged.

SPMI shows one beneficial inline:

         -63 (-12.73% of base) : 377923.dasm - System.DirectoryServices.Protocols.AddRequest:.ctor(System.String,System.String):this

[briansull]

If you gather PerfScore diffs on this chnage there shouldn't be much of a regression, as the block is marked as RarelyRun (with a weight of zero). The remaining diffs will be based upon codesize and will be greatly reduced:

emit.h:1321:#define PERFSCORE_CODESIZE_COST_HOT 0.10f
emit.h:1322:#define PERFSCORE_CODESIZE_COST_COLD 0.01f

However sometimes extra code in the Cold region can cause changes in register usage (extra callee saves, extra spills , etc..) in the hot regions and looking into those can often reveal area that we can improve upon.

[AndyAyersMS]

This change allows inlining of some methods with conditional throws. The throw block will be rarely run, but the rest of the inlined method won't be. So perf scores end up higher too.

Summary of Perf Score diffs:
(Lower is better)

Total PerfScoreUnits of diff: 84.47 (0.00% of base)
    diff is a regression.

Top file regressions (PerfScoreUnits):
       30.50 : System.Composition.AttributedModel.dasm (15.91% of base)
       21.30 : System.Security.Cryptography.Pkcs.dasm (0.02% of base)
       19.10 : System.Private.CoreLib.dasm (0.00% of base)
        7.80 : System.DirectoryServices.Protocols.dasm (0.02% of base)
        5.77 : System.Runtime.Numerics.dasm (0.02% of base)

5 total files with Perf Score differences (0 improved, 5 regressed), 160 unchanged.

Top method regressions (PerfScoreUnits):
       21.30 (10.89% of base) : System.Security.Cryptography.Pkcs.dasm - PkcsHelpers:DeepCopy(CmsRecipientCollection):CmsRecipientCollection
       19.10 ( 4.38% of base) : System.Private.CoreLib.dasm - CustomAttributeData:get_NamedArguments():IList`1:this
       17.10 (259.09% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor(String):this
       13.40 (203.03% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor():this
        7.80 ( 7.75% of base) : System.DirectoryServices.Protocols.dasm - AddRequest:.ctor(String,String):this
        5.77 ( 3.22% of base) : System.Runtime.Numerics.dasm - BigNumber:HexNumberToBigInteger(byref,byref):bool

Top method regressions (percentages):
       17.10 (259.09% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor(String):this
       13.40 (203.03% of base) : System.Composition.AttributedModel.dasm - SharedAttribute:.ctor():this
       21.30 (10.89% of base) : System.Security.Cryptography.Pkcs.dasm - PkcsHelpers:DeepCopy(CmsRecipientCollection):CmsRecipientCollection
        7.80 ( 7.75% of base) : System.DirectoryServices.Protocols.dasm - AddRequest:.ctor(String,String):this
       19.10 ( 4.38% of base) : System.Private.CoreLib.dasm - CustomAttributeData:get_NamedArguments():IList`1:this
        5.77 ( 3.22% of base) : System.Runtime.Numerics.dasm - BigNumber:HexNumberToBigInteger(byref,byref):bool

6 total methods with Perf Score differences (0 improved, 6 regressed), 244490 unchanged.

[sandreenko]

LGTM

[briansull]

LGTM