Implement guardrails for Fibonacci benchmark utilities

[arthurpaulino]

This PR implements an automatic computation of the linear and angular coefficients for the iteration count formula and uses that to assert that the hardcoded constants are correct. These assertions run in the beginning of the benchmark routine.

It also removes examples/fibonacci.rs as its main purpose was to validate the constants that we hardcoded. The function to extract the Fibonacci result was factored out to benches/common/fib.rs and is used for the tests mentioned above.

Closes #990

[huitseeker]

I really don't want to make fibonacci utilities part of the public API of Lurk.

If this was just for benches, a better approach here is to open another crate in our workspace (which already has several) , make it non-publishable, and make this crate a dev-dependency. Unfortunately, this would not work for examples.

One approach we could envision is to have an unpublished application crate in the workspace that contains essentially the code of the example (and example is morally a public external module anyway), and that crate could have a benchmark. This would work because both bench and example are supposed to only call into the lurk public functions, which is something they can do from an external crate.

This will require some CI support (/cc @samuelburnham for visibility) and has no downside I can see besides being quite a bit of work: perhaps not our first priority atm, to save a few lines of duplication.

[arthurpaulino]

Some details I'm not very happy about the latest commit.
Also, it needs the changes so CI doesn't break, since this moves benches/fibonacci.rs to lurk-fibonacci/benches/throughput.rs

@samuelburnham I need your help on this one

[huitseeker]

Some details I'm not very happy about the latest commit.

Right so in the grand scheme of things the question is: was this worth it? If we had a grand benchmark such that say the Lurk function is hard to maintain in two places, maybe. In our case, maybe the status quo is enough?

[arthurpaulino]

maybe the status quo is enough?

• The status quo doesn't have the tests test_coeffs and test_fib_io_matches which materialize @porcuquine's request
• The status quo also doesn't centralize the common functions

For the sake of correctness and simplicity, I'd rather revert the latest commit and have src/fib.rs. At least until we migrate away from it for throughput benchmarks

[huitseeker]

Excellent, thank you so much!