Make addition of CC algorithms modular

[huitema]

We like to use picoquic to investigate congestion control and test multiple algorithms. However, we do not want to load the code of multiple algorithms in every executable program that uses picoquic. For example, some developers are looking at Arduino class devices, and want to make the code as small as possible.

The general approach is to use dynamic linking. If an object file does not include any function required by the program, it will not be linked into the executable. But to achieve that, the program shall not have any reference to the function, including no calls in a rarely used branch, and no addition to the function pointer in a list of "external" objects. We apply this logic to congestion control algorithms as follow:

• remove the list of congestion algorithm pointers in picoquic.h
• declare a separate header file for each congestion control algorithm, to be included in the programs using that algorithm
• add a dynamic table of congestion control algorithm, either initialized by a program to the small set that it uses or initialized by a "register all algorithms" function for programs like picoquicdemo that want to surface all available options
• modify the "config" functions to use the table of algorithms registered by the program.

[huitema]

Review of PR #1853 show that we need one more effort to get a fully generic solution: some generic way to pass algorithm specific parameters from the UI/CLI/App to the CC code. Let's work on that before merging this PR.

[hfstco]

Just a note from my side:

Because there is a header file for each CC now, we can also consider to define the functions of the CC there. Currently there is no way to do some unit tests with the CC algos. And it should not increase the file size of the executable too.

But I think we could discuss and do that in a separate PR.

[huitema]

@hfstco Thanks for the review. Yes, adding unit tests for CC algos would be a great idea, and documenting the functions to be tested in the header files would be very neat. I am also adding in this PR a generic way to pass an option string to the CC Algo. This is tested in one of the BBR/Wi-Fi tests.

[huitema]

Next step after this is to update the picoquic_ns project to automatically pick up the definition of the algorithms that we add to the picoquic depot. With that, we should be able to quickly assess new protocols like Hybla...