Move solve and unify into their own crates

compiler/solve/Cargo.toml

@@ -0,0 +1,20 @@
+[package]
+name = "roc_solve"
+version = "0.1.0"
+authors = ["Richard Feldman <oss@rtfeldman.com>"]
+edition = "2018"
+
+[dependencies]
+roc_collections = { path = "../collections" }
+roc_region = { path = "../region" }
+roc_module = { path = "../module" }
+roc_types = { path = "../types" }
+roc_can = { path = "../can" }
+roc_unify = { path = "../unify" }
+
+[dev-dependencies]
+pretty_assertions = "0.5.1 "
+maplit = "1.0.1"
+indoc = "0.3.3"
+quickcheck = "0.8"
+quickcheck_macros = "0.8"

compiler/solve/src/lib.rs

@@ -0,0 +1,14 @@
+#![warn(clippy::all, clippy::dbg_macro)]
+// I'm skeptical that clippy:large_enum_variant is a good lint to have globally enabled.
+//
+// It warns about a performance problem where the only quick remediation is
+// to allocate more on the heap, which has lots of tradeoffs - including making it
+// long-term unclear which allocations *need* to happen for compilation's sake
+// (e.g. recursive structures) versus those which were only added to appease clippy.
+//
+// Effectively optimizing data struture memory layout isn't a quick fix,
+// and encouraging shortcuts here creates bad incentives. I would rather temporarily
+// re-enable this when working on performance optimizations than have it block PRs.
+#![allow(clippy::large_enum_variant)]
+
+pub mod solve;