Refactor couplings tests and benchmark

[alecandido]

@niclaurenti there is a huge copy-pasting of tests in https://github.com/NNPDF/eko/blob/tempdir/tests/eko/test_couplings.py

Essentially, all the _QED and _mix ones are the same code, but for order. Please, take the time to refactor it, because it only makes it harder to maintain (I had to repeat the same changes over and over every time that something changes in couplings).

Originally posted by @alecandido in #172 (comment)

[alecandido]

Actually, also isolated benchmarks should be refactored in a similar way @felixhekhorn

I accepted the repeated code in the tests some time ago, but that was an error: tests code is still code that has to maintained, so we should develop as a package on its own, factorizing as much as possible.

The https://github.com/NNPDF/eko/blob/3f81e7b673c636e3b74f7f11d353e5c055c748c4/benchmarks/eko/benchmark_strong_coupling.py file contains 12 tests, most of them are making repeated use of the same pattern(s). Fixing them is unreasonably expensive.

Similar situation for other tests and isolated benchmarks, while benchmarks themselves are factorized. At this point, we can also use banana to collect common fixtures and repeated patterns...
But at least we should refactor each test/isobench module on its own.
I will open a dedicated issue for this part, this is specifically dedicated to couplings.

Another important point is that we should drop the current:

eko/benchmarks/eko/benchmark_strong_coupling.py

Lines 14 to 46 in 3f81e7b

	try:
	import lhapdf
	use_LHAPDF = True
	except ImportError:
	use_LHAPDF = False
	# try to load APFEL - if not available, we'll use the cached values
	try:
	import apfel
	use_APFEL = True
	except ImportError:
	use_APFEL = False
	# try to load pegasus - if not available, we'll use the cached values
	try:
	import pegasus
	use_PEGASUS = True
	except ImportError:
	use_PEGASUS = False
	def ref_couplings(
	ref_values,
	ref_scale: float,
	) -> types.CouplingsRef:
	return types.CouplingsRef(
	alphas=types.FloatRef(value=ref_values[0], scale=ref_scale),
	alphaem=types.FloatRef(value=ref_values[1], scale=nan),
	)

Isolated benchmarks are designed to run in the dedicated container, so there is APFEL, there is LHAPDF, and there is PEGASUS.
If needed for development we can add a Dev Container to the repo https://containers.dev/, that can be simply used from VS code as well, and they have support for specification of base image (and we can use the benchmark one). In this way it should simplify the workflow also for developers.

[felixhekhorn]

Tests are updated in #187

[alecandido]

Tests are updated in #187

Perfect, thanks :D

I would keep the issue for the benchmarks :)