(Improvements) Ring-buffer and XdrReaderWriter optimizations, assorted Span/Memory method additions

[pl752]

Decided to split up my previous pull request (#1247) to multiple PRs due to it having set of changes too broad in my opinion. Ran tests as usual for fb3 server and embedded. Updated to current master branch state. Also decided to throw in reworked ring-buffer Queue's to be more large object friendly.

TLDR: Ellimination of most of the repeated allocations for intermediate conversions, also added inplace copy methods for I/O and reworked intermediate network buffers to use proper ring-buffer resizable arrays with unitary copying instead of per-byte queueing, massively boosting large objects throughput.

Practical benchmarks (3 int and 1 char100 columns, real I/O with fb3 server on localhost and nvme)

Master:

Method	Rows	Mean	Error	StdDev	Median	Gen0	Gen1	Allocated
SelectAndMap_Main_ReusedBufferAsync	10	1.796 ms	0.0649 ms	0.1904 ms	1.784 ms	-	-	457.27 KB
SelectAndMap_Main_ReusedBufferAsync	100	8.326 ms	0.5724 ms	1.5955 ms	7.472 ms	-	-	4502.13 KB
SelectAndMap_Main_ReusedBufferAsync	1000	31.037 ms	3.7512 ms	10.8231 ms	24.973 ms	5000.0000	1000.0000	44985.72 KB
SelectAndMap_Main_ReusedBufferAsync	10000	337.278 ms	16.5461 ms	48.7865 ms	334.254 ms	55000.0000	10000.0000	449543.01 KB
SelectAndMap_Main_ReusedBufferAsync	100000	3,467.848 ms	98.9785 ms	288.7249 ms	3,435.619 ms	550000.0000	114000.0000	4494685.62 KB

New (isolated from Rune opt (#1252)):

Method	Rows	Mean	Error	StdDev	Median	Gen0	Gen1	Allocated
SelectAndMap_Main_ReusedBufferAsync	10	1.614 ms	0.0500 ms	0.1450 ms	1.584 ms	-	-	447 KB
SelectAndMap_Main_ReusedBufferAsync	100	9.760 ms	0.1127 ms	0.0880 ms	9.746 ms	-	-	4411.49 KB
SelectAndMap_Main_ReusedBufferAsync	1000	22.283 ms	2.1941 ms	6.1524 ms	19.432 ms	5000.0000	1000.0000	43999.39 KB
SelectAndMap_Main_ReusedBufferAsync	10000	221.000 ms	9.0103 ms	26.1406 ms	221.383 ms	53000.0000	10000.0000	439534.85 KB
SelectAndMap_Main_ReusedBufferAsync	100000	2,317.527 ms	46.2359 ms	108.0750 ms	2,342.930 ms	538000.0000	108000.0000	4394711.29 KB

Combined with #1252:

Method	Rows	Mean	Error	StdDev	Median	Gen0	Gen1	Allocated
SelectAndMap_Main_ReusedBufferAsync	10	471.8 us	15.22 us	43.68 us	462.0 us	-	-	43.98 KB
SelectAndMap_Main_ReusedBufferAsync	100	1,638.7 us	62.67 us	182.81 us	1,590.5 us	-	-	383.96 KB
SelectAndMap_Main_ReusedBufferAsync	1000	13,283.3 us	2,299.73 us	6,561.26 us	9,672.6 us	-	-	3785.55 KB
SelectAndMap_Main_ReusedBufferAsync	10000	92,663.6 us	2,673.87 us	7,841.99 us	91,372.1 us	4000.0000	-	37583.41 KB
SelectAndMap_Main_ReusedBufferAsync	100000	897,179.0 us	17,724.27 us	24,846.95 us	900,061.5 us	45000.0000	5000.0000	375392.09 KB

Perf (defaults):

Method	Job	BuildConfiguration	DataType	Count	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
Execute	NuGet	ReleaseNuGet	bigint	100	19,426.6 us	195.03 us	172.89 us	1.00	0.01	31.2500	290.31 KB	1.00
Execute	Project	Release	bigint	100	19,731.2 us	376.18 us	369.46 us	1.02	0.02	-	192.32 KB	0.66
Fetch	NuGet	ReleaseNuGet	bigint	100	461.9 us	5.48 us	5.12 us	1.00	0.02	5.8594	52.02 KB	1.00
Fetch	Project	Release	bigint	100	449.6 us	3.00 us	2.81 us	0.97	0.01	3.9063	39.03 KB	0.75
Execute	NuGet	ReleaseNuGet	varch(...) utf8 [30]	100	19,592.1 us	172.88 us	144.36 us	1.00	0.01	31.2500	294.24 KB	1.00
Execute	Project	Release	varch(...) utf8 [30]	100	19,419.8 us	96.03 us	80.19 us	0.99	0.01	-	194.64 KB	0.66
Fetch	NuGet	ReleaseNuGet	varch(...) utf8 [30]	100	466.2 us	2.37 us	2.22 us	1.00	0.01	6.8359	55.9 KB	1.00
Fetch	Project	Release	varch(...) utf8 [30]	100	458.4 us	3.01 us	2.67 us	0.98	0.01	3.9063	39.77 KB	0.71

Boost for large object I/O is significant. Benchmarks were performed without compression and encryption (typical for walled LAN / server sharing db with apps in single env). For small data types and writing operations, changes in practice are much smaller, but some difference in allocations can be observed still. Such an effect on performance is mostly due to ellimination of copies and ability for system to use packed operations (r/w with 64 bit ops instead of 8x8 bit ops) or even simd when dealing with the buffers, and also jit should be much happier with the new installment. For large blobs with the same mode boost can be even bigger with the bypass mode (less copies).