Fix `Grid::convolute` caching strategy

[cschwan]

Right now Grid::convolute always has two_caches = true (I've discovered this while running the code in rust-gdb),

https://github.com/N3PDF/pineappl/blob/6ccb9a49fa9d3baa06f9958fd1db6f9286983449/pineappl/src/grid.rs#L343

which misses a few optimizations whenever the two initial-state hadron are the same.

[cschwan]

Plan:

• replace arguments xfx1, xfx2 and alphas with a single object of PdfCache, implemented in commit c6c5a0b,
• which takes care of caching the arguments of every PDF,
• in such a way that pp, ppbar, etc. rely on a single PDF (also in commit c6c5a0b).

[cschwan]

Here a benchmark, using the grids from NNPDF/pineapplgrids@6960247, running the following command on a computer with an Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz:

time for i in *.pineappl.lz4; do pineappl --silence-lhapdf convolute $i NNPDF40_nnlo_as_01180; done > output

1. Using pineappl from commit 6911055 takes 13 seconds, and this suffers from the problem that both initial states have a separate cache, meaning the caching is imperfect. However, due to symmetrization it should be almost perfect in practice.
2. Using pineappl from commit c6c5a0b, with convolute2 instead of convolute in the helper function it takes 73 seconds for the same command; this is because convolute doesn't use caching at all at this stage.

[cschwan]

A similar benchmark with pdf_uncertainty instead of convolute takes 1m37s in the first case and 4m9s in the second case; this subcommand uses 8 cores. This just goes to show that PDF caching is still very much needed.

[cschwan]

More work done in commits f2d2e0e, 5b81a3c, 66b38fc and 4c931ef.

[cschwan]

Commit aa3c4b7 implements the actual caching; with it convolute and convolute2 are basically equally fast.

[cschwan]

The new cache is now activated in commit cedd348 for the CLI.

[cschwan]

The only remaining item is to convert Grid::convolute_subgrid.

[cschwan]

Commit 4dbedc6 adds the remaining changes.