L3-Performance benchmarking: Maintain main-perf dev-branch.

[gapisback]

We discussed the need to do controlled performance benchmarking to measure the overheads of L3-logging in comparison to other schemes such as:

1. Baseline: With no L3 or any other form of logging.
2. Logging of messages using fprintf() to some output file, with or w/o fflush()
3. Logging using the header-only spdlog package
4. Different implement of L3-APIs where, say, l3_init() falls-back to fopen() specified file. And then l3_log_simple() defaults to a simple fprintf()
5. Extending recently-created client-server-msgs-perf performance exerciser to driver micro-benchmarking performance.

---

I think trying to maintain all scripts, drivers to do these kinds of performance-benchmarking experiments in core L3 repo-sources will become difficult.

Proposal

1. Create a dev-branch called main-perf-bm which will be cut-off of /main.
2. All specialized perf-specific code, e.g. conditinonaly compiled code in client-server-msgs-perf sources will be implemented in this branch.
3. We can maintain a submodule of spdlog repo in this dev-branch, to simplify integration of that library withclient-server-msgs-perf programs.
4. Any future perf-benchmarking specific sources / tools can be developed and maintained in just this one branch, which likely goes thru its own variation of CI-builds (tbd).

@gregthelaw -- Let me know if this scheme works for you and any issues you see with this scheme.

If agreeable, I can start working on this dev-branch to start building out the perf-bm tooling.

[gregthelaw]

@gregthelaw -- Let me know if this scheme works for you and any issues you see with this scheme.

sounds good!

[gapisback]

Proposal has been accepted and moved over to new issue 45.