Optimize dispose logic

[bitfaster]

Experiment - can JIT elide code like:

if (value is IDisposable d)
{
   d.Dispose();
}

[bitfaster]

Benchmark is bogus (ran a minute ago and not optimized was fastest):

Method	Mean	Error	StdDev	Ratio	RatioSD
HandWritten	285.6 ns	2.21 ns	1.96 ns	1.00	0.00
NotOptimized	287.8 ns	3.83 ns	3.58 ns	1.01	0.02
GenericDisposer3	285.9 ns	4.24 ns	3.76 ns	1.00	0.01

[bitfaster]

Stupid :) - this is checking if Type is IDisposable, because the result of typeof(T) is an instance of Type.

[bitfaster]

can try using static readonly field - issue 4920, but looks like it doesn't apply to generics.

[bitfaster]

Wrapping the standard if value is IDisposable check into a generic is equal speed/code size:

Method	Mean	Ratio	Code Size	Allocated
HandWritten	3.527 us	1.00	0 KB	23 KB
NotOptimized	3.505 us	0.99	0 KB	23 KB
GenericDisposerReadonlyProperty	8.293 us	2.36	0 KB	47 KB
GenericDisposerStdCheck	3.502 us	0.99	0 KB	23 KB

And it looks like it does decrease the size of the code by about 4% (from running the LRU bench with DisassemblyDiagnoser):

With Disposer:

Method	Code Size
ConcurrentDictionary	340 B
FastConcurrentLru	451 B
ConcurrentLru	473 B
FastConcurrentTLru	638 B
ConcurrentTLru	709 B
ClassicLru	738 B

Original:

Method	Code Size
ConcurrentDictionary	340 B
FastConcurrentLru	473 B
ConcurrentLru	495 B
FastConcurrentTLru	666 B
ConcurrentTLru	737 B
ClassicLru	756 B

The actual runtime is pretty much identical - within the variance from running the same thing over and over again.

[bitfaster]

Generated assembly is the same as hand written code:

; BitFaster.Caching.Benchmarks.DisposerBench.HandWritten()
       push      rsi
       sub       rsp,20
       xor       esi,esi
M00_L00:
       mov       rcx,offset MT_BitFaster.Caching.Benchmarks.Disposable
       call      CORINFO_HELP_NEWSFAST
       cmp       byte ptr [rax+8],0
       jne       short M00_L01
       mov       byte ptr [rax+8],1
M00_L01:
       inc       esi
       cmp       esi,3E8
       jl        short M00_L00
       add       rsp,20
       pop       rsi
       ret
; Total bytes of code 48

; BitFaster.Caching.Benchmarks.DisposerBench.GenericDisposerStdCheck()
       push      rsi
       sub       rsp,20
       xor       esi,esi
M00_L00:
       mov       rcx,offset MT_BitFaster.Caching.Benchmarks.Disposable
       call      CORINFO_HELP_NEWSFAST
       cmp       byte ptr [rax+8],0
       jne       short M00_L01
       mov       byte ptr [rax+8],1
M00_L01:
       inc       esi
       cmp       esi,3E8
       jl        short M00_L00
       add       rsp,20
       pop       rsi
       ret
; Total bytes of code 48