Channels optimization

[flcl42]

Is your feature request related to a problem? Please describe.
Channels are used to efficiently pass data through the protocol stack, they are queues of byte chunks and copy happens only when API does not allow working with ReadOnlySequence. Still they need several improvements:

• initially allocate data in consistent way
• de-allocate data when it's read
• when the data is being materialized, provide access to underlying spans, to cover more use cases?
• detect leaks and de-allocate data that will not be read for sure

Describe the solution you'd like

• Measure current memory allocations and determine the current level of efficiency, if it's worth it:
• Add allocator to be used by transport protocols
• Improve read api
• De-allocate when needed

Describe alternatives you've considered
TBD

Additional context
How it's now:

• in TCP we simply allocate bytes and hope for GC, can we do it in more controlled way?: https://github.com/NethermindEth/dotnet-libp2p/blob/main/src/libp2p/Libp2p.Protocols.IpTcp/IpTcpProtocol.cs#L85
• in QUIC it's the same https://github.com/NethermindEth/dotnet-libp2p/blob/main/src/libp2p/Libp2p.Protocols.Quic/QuicProtocol.cs#L267
• the data is read, encrypted and then sent, read data is now can be freed: https://github.com/NethermindEth/dotnet-libp2p/blob/main/src/libp2p/Libp2p.Protocols.Noise/NoiseProtocol.cs#L171