API refinement: make RX pipeline zero-copy as a side effect of the OOO reassembly support

[pavel-kirienko]

Close #36

Add RX frame parser (a simple function) with tests.

Update the RX pipeline as described in https://forum.opencyphal.org/t/refined-requirements-for-the-rx-pipeline-in-libudpard/1938.

Pros:

• Zero-copy RX pipeline

Cons:

• The udpardRx*Receive functions take ownership of the payload buffer.
• The memory usage pattern depends on the MTU of the transmitting node even for the same extent. Small MTU --> more payload fragments --> more memory fragments to keep, and vice versa.

The docs are a bit rough around the edges but the idea should be clear. One missing feature is a trivial helper function that takes payload fragments and returns a contiguous chunk of memory instead (this is expensive both memory-wise and time-wise).

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
2 Code Smells

86.0% Coverage
0.0% Duplication

[lydia-at-amazon]

Cons:

The udpardRx*Receive functions take ownership of the payload buffer.

Just want to document what we talked about in the dev call, the zero-copy design assumes that udpard can directly take over ownership of the buffer from the network stack. A lot of microcontrollers use DMA to store the payload buffer in a section of memory that udpard would not be able to take ownership of and would require making a copy. Are we going to provide an API to allow for this case?

[lydia-at-amazon]

One missing feature is a trivial helper function that takes payload fragments and returns a contiguous chunk of memory instead (this is expensive both memory-wise and time-wise).

Do we have a ticket for this? I think almost all applications are going to want the contiguous chunk of memory and not the fragments, so we're kind of feature incomplete without this function.

[pavel-kirienko]

Are we going to provide an API to allow for this case?

The plan is to let the application copy the data from the memory managed by the NIC driver into a new buffer and move it into the library. This solution does not require new API. We can optionally provide convenience functions for this later.