refactor: Performance tuning (stream + incr experiments)

[DaveSkender]

Side piece of

• feat: Full chain streaming prototypes #1092

• See inline to dos and deferrals to log

[DaveSkender]

Starting point:

Method	Mean	Error	StdDev
AdlHub	30.87 us	5.544 us	0.304 us
AtrHub	45.33 us	20.505 us	1.124 us
AtrStopHub	97.21 us	18.485 us	1.013 us
AlligatorHub	50.89 us	16.643 us	0.912 us
EmaHub	26.27 us	8.360 us	0.458 us
QuoteHub	13.02 us	1.225 us	0.067 us
QuotePartHub	29.28 us	22.817 us	1.251 us
RenkoHub	159.02 us	18.238 us	1.000 us
SmaHub	36.62 us	9.782 us	0.536 us
TrHub	35.04 us	9.639 us	0.528 us

[DaveSkender]

Method	Mean	Error	StdDev
AtrHub	45.73 us	1.910 us	0.105 us
AtrStopHub	72.14 us	4.821 us	0.264 us
TrHub	35.92 us	5.774 us	0.316 us

[DaveSkender]

• Compare GetAlligator() to main, possibly regressed perf test
• Compare difference between EMA and TR, TR should about the same
• Try rolling buffer for series, possibly a resetting variant for hubs, notably for SMA-types
• Compare without Reusable base class, or just as abstract.

[DaveSkender]

New incremental “buffered” list types

Method	Mean	Error	StdDev
EmaListQ	7.194 us	0.0750 us	0.0664 us
EmaListP	6.634 us	0.0737 us	0.0653 us
EmaArray	5.206 us	0.0999 us	0.0934 us
EmaSeries	8.430 us	0.0891 us	0.0834 us

[DaveSkender]

This is a better equivalence test for incremental "buffer" methods for EMA.
Static series are still the most performant overall.
The "Inc" methods are only faster than the original because we're not sorting the input, but are equivalently 4× slower, likely due to re-writes from the RemoveAt(0) buffer management.

Method	Mean	Ratio	Notes
IncRus	7.35us	2.60	Incrementing with IReusable input value, one at a time.
IncRusBatch	7.27us	2.58	Incrementing with IReusable batch.
IncQot	7.42us	2.63	Incrementing with IQuote input value, one at a time.
IncQotBatch	7.29us	2.58	Incrementing with IQuote batch.
IncPrm	5.15us	1.83	Incrementing with primitive double, one at a time.
IncPrmBatch	5.27us	1.87	Incrementing with primitive double batch.
SeriesEqiv	2.82us	1.00	Original series without sorting (benchmark).
SeriesOrig	8.57us	3.04	Original series with sorting.
IncremEqiv	13.06us	4.63	Incrementing with Quote batch and sorting, for a same-same comparison.
StreamEqiv	27.32us	9.67	Stream hub. It has more self-healing and analysis features (more overhead).

Still need to fix/try out a few scenarios:

• Add argument null checks
• Try Add overload syntax again; this was conflicting with base List.Add(double?).
• Try rename class to "Ema" again, there were some conflicts with utilities.
• Add buffer to original series, just for a sanity check
• Explore alternatives to RemoveAt(0) for incrementing buffer.
• Consider removing primitive variant since it's an old model, like TA-LIB, that just offloads work to the user. Or, defer to the backlog.

Also, given the 3× improvement without sorting:

• Consider offloading sorting responsibility for time-series methods to the user. I suspect most batch quote providers will or can be fetched in a correct sort anyway.
  • Related: consider using IReadOnlyList<IQuote> quotes as input, instead of IEnumerable<IQuote> quotes
  • Update documentation accordingly
• Expand Validate() to have generic overload (related)?

Increment variants deferred to:

• add standalone buffer-style increment methods #1091

[DaveSkender]

• Check if Reusable can be contravariant to TReusable, or other in/out type keywords
  • apply to AlligatorHub and other HL2 scenarios to avoid excess casting overhead
• Check if more interfaces can be used for input parameters, instead of concrete types, if performant
• add notnull constraint to ISeries and remove incremental types null reference checks, if possible (may not work), seek other similar fixes

[DaveSkender]

• test for CMO isUp works as expected when there’s no price change

[DaveSkender]

Minor improvements to QuoteHub and RenkoHub

Method	Mean	Error	StdDev
AdlHub	34.411 μs	21.475 μs	1.1771 μs
AtrHub	41.800 μs	4.962 μs	0.2720 μs
AtrStopHub	77.144 μs	33.461 μs	1.8341 μs
AlligatorHub	48.238 μs	3.007 μs	0.1648 μs
EmaHub	28.427 μs	8.953 μs	0.4907 μs
QuoteHub	8.338 μs	1.511 μs	0.0828 μs
QuotePartHub	27.229 μs	11.569 μs	0.6341 μs
RenkoHub	40.373 μs	8.937 μs	0.4899 μs
SmaHub	36.289 μs	25.591 μs	1.4027 μs
TrHub	32.927 μs	7.490 μs	0.4106 μs

[DaveSkender]

Method	Mean	Error	StdDev
AdlHub	34.356 us	4.0523 us	0.2221 us
AtrHub	42.367 us	11.7349 us	0.6432 us
AtrStopHub	70.069 us	1.8487 us	0.1013 us
AlligatorHub	47.881 us	5.5744 us	0.3056 us
EmaHub	26.778 us	2.4779 us	0.1358 us
EmaHub2	26.453 us	3.7542 us	0.2058 us
QuoteHub	8.293 us	2.0734 us	0.1137 us
QuotePartHub	27.764 us	4.2852 us	0.2349 us
RenkoHub	39.993 us	0.8697 us	0.0477 us
SmaHub	37.197 us	6.0623 us	0.3323 us
TrHub	33.422 us	10.2533 us	0.5620 us