Bump Iris version to match LambdaRust-Coq.

See merge request !4

- evmap is dropped
- per-item invariant is implemented with inv-in-inv
- peritem.v is simplified by proving an ad-hoc iter spec
- update to latest iris
- related fixes

Improve build system & Update dependency

See merge request !3

Iris as submodule; CI support

This adds Iris as a submodule, matching the structure in https://gitlab.mpi-sws.org/FP/LambdaRust-coq/. Furthermore, it sets up the CI so after pushing, the repo is built against the given Iris version (with a cache to avoid needless rebuilding of Iris).

@zhangz if you have any unpushed changes, please push them. This branch moves all files to a subfolder `theories/`, so it will conflict with any other change.

See merge request !2

Simpler syncer specs

This simplifies the syncer specs.
The sequential syncer only is slightly improved, by changing
```
  Definition synced R (f' f: val) :=
    (□ ∀ P Q (x: val), ({{ R ★ P x }} f x {{ v, R ★ Q x v }}) →
                       ({{ P x }} f' x {{ v, Q x v }}))%I.
```
such that `P`, `Q` don't factor in a dependency on `x` any more. Since all three are quantified at the same time, the two specs are equivalent -- and the one where `P` is not a predicate and `Q` is exactly the postcondition is arguably simpler:
```
  Definition synced R (f f': val) :=
    (□ ∀ P Q (x: val), ({{ R ★ P }} f x {{ v, R ★ Q v }}) →
                       ({{ P }} f' x {{ Q }}))%I.
```

For the atomic spec, the changes are deeper. The currying on some location `l` is entirely removed. `is_atomic_syncer` better mirrors `is_syncer` from the sequential version. And finally, we use "original" atomic triples and not some modified version thereof (so `atomic_triple_base` is killed).

TODO: Fix `atomic_pcas.v`, it relied on the location currying done by `atomic_sync.v`. However, I don't entirely grok that file. @zhangz could you give that a look?

See merge request !1

* The currying on some location l is entirely removed.
* is_atomic_syncer better mirrors is_syncer from the sequential version.
* We use "original" atomic triples and not some modified version thereof (so atomic_triple_base is killed).
* Don't embed the Iris turnstile into Iris.

requires a new Iris version