[fleche] Caching / sharing of .vo contents

[ejgallego]

We add a basic .vo parsing cache; this often results in large memory savings when working simultaneously with several files.

Based on work from #641 and coq/coq#17393

[ejgallego]

@Alizter memory use (approx) with patch, opening a lot of files in HoTT:

• with cache=true: 84Mw
• with cache=false: 144Mw

So while not the panacea, the reduction in memory use seems quite useful, in particular for devs with large .vo size.

I'm very interested in seeing your numbers.

[ejgallego]

cc: @SkySkimmer @ppedrot in case you are curious about the kind of improvements we see by sharing .vo files (in stdlib that amounts to 8% faster compilation)

[ejgallego]

Loading all of HoTT in the same server: 1169 Mw to 668 Mw , quite a considerable saving @Alizter