GitLab

- The typeclass `PureExec e1 e2` controls pure deterministic
  reductions
- General tactics rel_pure_l and rel_pure_r that depend on that typeclass
- The same typeclass can be used for WP tactics and tp tactics, potentially

- Use the type of literals in `val`
- Notation for `match`
- "Better" coercions

- Add more binary ops to the language
- An example refinement for module types

Modify the logical relation judment to include an environment Δ that
contains a list of semantic types to interpret free type variables.

- Change the types in the examples slightly
- Use notations in the examples
- Modify some tactics to make the proofs more smooth

- This allows us to get rid of `rewrite !Closed_subst_id.`

- Twiggle with the burden of proof in log_related_arrow
- rel_rec_l/r returns a goal with an (unsimplified) lamsubst
  and can detect locked lambdas
- Fix some notational issues [hax.v]
- Rewrite the counter refinement proof to make greater use of lamsubst

- rename `bin_log_related_bind_l` to `bin_log_related_wp_l`
- use `bin_log_related_wp_l` to simplify derived rules

Using the lambdasubst hack-- instead of formulating
lemmas for (App e1 v), formulate them for (e1[x:=v]).

..by introducing a hint database for typeability

- Make sure that arguments to the lemmas are in the consistent order:
\Gamma, E1, E2, K, e1, e2, \tau -- from left to right, as you write it
 on paper
- Add masks to all the lemmas
- Add missing rules
- Remove unnecessary closedness conditions

- Remove commented out code
- Pull std++ related lemmas into a separate file

The current prove does not reuse certain lemmas

We are switich away from De Bruijn indices because
1) it's hard to read terms with De Bruijn indices
2) they are making lots of things slower

The new _with_lock rule does not mentioned lower level details of the
relational interpretation.