Optimize ConcurrentLru read throughput (volatile)

[bitfaster]

Improve read throughput by ~150% by avoiding volatile writes (store-load memory barrier).

Before

After

TODO: investigate read latency/code size. I would assume impact is negligible and this is a good tradeoff.

[coveralls]

coverage: 99.218% (+0.07%) from 99.149%
when pulling 3f8eef4 on users/alexpeck/fastmark
into aeae236 on main.

[bitfaster]

Method	Runtime	Mean	StdDev	Ratio	Code Size
ConcurrentDictionary	.NET 6.0	7.549 ns	0.0378 ns	1.00	NA
FastConcurrentLru	.NET 6.0	9.893 ns	0.1486 ns	1.31	7,124 B
ConcurrentLru	.NET 6.0	15.437 ns	0.0481 ns	2.04	7,371 B
AtomicFastLru	.NET 6.0	27.798 ns	0.1106 ns	3.68	NA
FastConcurrentTLru	.NET 6.0	11.911 ns	0.0872 ns	1.58	7,560 B
FastConcLruAfterAccess	.NET 6.0	12.419 ns	0.0656 ns	1.65	6,831 B
FastConcLruAfter	.NET 6.0	14.761 ns	0.1330 ns	1.96	8,167 B
ConcurrentTLru	.NET 6.0	21.596 ns	0.2259 ns	2.86	7,836 B
ConcurrentLfu	.NET 6.0	29.011 ns	0.4737 ns	3.84	NA
ClassicLru	.NET 6.0	44.402 ns	0.8194 ns	5.88	NA
RuntimeMemoryCacheGet	.NET 6.0	116.210 ns	1.2554 ns	15.39	89 B
ExtensionsMemoryCacheGet	.NET 6.0	49.176 ns	0.6670 ns	6.51	119 B
ConcurrentDictionary	.NET Framework 4.8	13.933 ns	0.1054 ns	1.00	4,127 B
FastConcurrentLru	.NET Framework 4.8	16.670 ns	0.1470 ns	1.20	27,400 B
ConcurrentLru	.NET Framework 4.8	20.384 ns	0.0891 ns	1.46	27,704 B
AtomicFastLru	.NET Framework 4.8	37.778 ns	0.1741 ns	2.71	358 B
FastConcurrentTLru	.NET Framework 4.8	28.804 ns	0.2912 ns	2.07	27,584 B
FastConcLruAfterAccess	.NET Framework 4.8	30.078 ns	0.1984 ns	2.16	358 B
FastConcLruAfter	.NET Framework 4.8	33.320 ns	0.6004 ns	2.39	358 B
ConcurrentTLru	.NET Framework 4.8	33.120 ns	0.1406 ns	2.38	27,936 B
ConcurrentLfu	.NET Framework 4.8	52.242 ns	0.6760 ns	3.75	NA
ClassicLru	.NET Framework 4.8	52.825 ns	0.2602 ns	3.79	NA
RuntimeMemoryCacheGet	.NET Framework 4.8	309.169 ns	3.1850 ns	22.19	79 B
ExtensionsMemoryCacheGet	.NET Framework 4.8	95.930 ns	0.8040 ns	6.89	129 B

[bitfaster]

This adds 4 instructions with 1 jump: