Generalize Uniqueness Analysis to arbitrary semirings

[anfelor]

This PR contains a proposed refactoring of uniqueness_analysis.ml. I have added it at @riaqn's request so that we don't lose track of this code. Ignore most of the diff: all the relevant changes are to uniqueness_analysis.ml in commits 255f533 and 4d69a80.

The refactoring reimagines the uniqueness analysis as a data flow analysis over arbitrary semirings. That makes it potentially much more useful in the future as any data that attaches to memory locations and is semiring-shaped can be collected in this analysis. Here is an overview over the changes:

• I remove the usage information from Usage_tree, Usage_forest and Paths. Instead, these are now generic datastructures that can hold arbitrary elements. I push them to the top of the file since the concept of "memory location mapped to data" can be understood without thinking about usages at all.
• I introduce a general functor for semirings and for "promotable" semirings that can live in trees. You can form the product of such semirings. Trees and forests over a promotable semiring are again semirings.
• The code for usages is largely unchanged. There is an unrelated change adding an access_order to errors for better error messages, but other than that, the only change is that I make Usage a promotable semiring.
• There are modules Tag, Learned_tags and Overwrites which are related to the overwriting extension and can be ignored. These are promotable semirings as well.
• I use two different semirings for patterns and expressions. Patterns yield a PF.t which is a forest semiring that contains usages and learned tags, while expressions yield a UF.t which is a forest semiring that contains usages and overwrites.This makes it type-safe to combine learned tags with overwrites (split_pf, learn_tags).
• The rest of the file only contains changes that follow directly from the design above.