SwiftQUIC: PERF: Avoid per-path scan unless round trip time has elapsed

[agnosticdev]

Each call to updateMaximumUnreadInboundBytesAllowed does a per-path scan on each inbound packet to tune the receive buffer.
This change gates the per-path scan behind a full round trip elapsing. This is based on the smoothedRTT of the current path.

Top of tree:

98.93 M 52.6%		  QUICStreamInstance.updateMaximumUnreadInboundBytesAllowed(dataLengthAdded:connection:)
84.00 M 44.7%	-      	specialized ManyToManyProtocolHandler.applyToAllPaths(_:)
44.00 M 23.4%	-      	  specialized Sequence.forEach(_:)
40.00 M 21.3%	-      	  protocol witness for ManyToManyProtocolHandler.multiplexingPaths.getter in conformance QUICConnection
2.00 M  1.1%	-	    FlowControlGlobals.shared.unsafeMutableAddressor
1.00 M  0.5%	- 	    QUICStreamInstance.computeReceiveHighWaterMarkIncrease(dataLength:rtt:mss:now:)

With this change:

42.62 M 48.1%	-      	QUICStreamInstance.updateMaximumUnreadInboundBytesAllowed(dataLengthAdded:connection:)
42.62 M 48.1%	-	      QUICStreamInstance.computeReceiveHighWaterMarkIncrease(dataLength:connection:now:)
35.62 M 40.2%	-      	   QUICPath.smoothedRTT.getter
22.00 M 24.8%	-	        swift_beginAccess	
8.19 M  9.2%	-	        <deduplicated_symbol>	
5.42 M  6.1%	- 	        DYLD-STUB$$swift_beginAccess

Savings of 56 megacycles for both client and server so 112 megacycles total.

[agnosticdev]

Note that the build issue on nightly-main is a known issue right now.

[tfpauly]

The previous code was getting the largest MSS and RTT across all paths. The newer code looks like it just looks at the current path. Is that a problem?

[rpaulo]

I asked @agnosticdev to remove it because this left over code (MPQUIC) which we don't need right now.

[agnosticdev]

Yes, we do not need it right now.