Enable incremental compilation in bazel rules

[johnynek]

This uses the work in #54 to incrementally compile bosatsu_library and avoid typechecking dependencies for given targets. bosatsu_test now reuses the library parsing and typechecking.

this uncovered the bugs in #279

[codecov-io]

Codecov Report

Merging #281 into master will decrease coverage by 0.36%.
The diff coverage is 20%.

@@            Coverage Diff             @@
##           master     #281      +/-   ##
==========================================
- Coverage   86.83%   86.47%   -0.37%     
==========================================
  Files          56       56              
  Lines        4536     4570      +34     
  Branches      181      195      +14     
==========================================
+ Hits         3939     3952      +13     
- Misses        597      618      +21

Impacted Files	Coverage Δ
core/src/main/scala/org/bykn/bosatsu/Program.scala	98.07% <ø> (+13.07%)	⬆️
...e/src/main/scala/org/bykn/bosatsu/PackageMap.scala	70.16% <0%> (-0.39%)	⬇️
cli/src/main/scala/org/bykn/bosatsu/Main.scala	0% <0%> (ø)	⬆️
core/src/main/scala/org/bykn/bosatsu/Predef.scala	93.75% <100%> (+0.04%)	⬆️
...main/scala/org/bykn/bosatsu/TypedExprToProto.scala	90.04% <71.42%> (-0.58%)	⬇️
.../src/main/scala/org/bykn/bosatsu/Declaration.scala	97.94% <0%> (-0.42%)	⬇️
...rc/main/scala/org/bykn/bosatsu/TotalityCheck.scala	96.64% <0%> (+0.37%)	⬆️
...e/src/main/scala/org/bykn/bosatsu/StringUtil.scala	98.33% <0%> (+1.66%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a392274...6e04eb0. Read the comment docs.

[johnynek]

note, this coverage decline is actually false, we exercise the Main.scala code in the bazel rules, but we don't merge that coverage into the main reports.

[johnynek]

going to merge this when it goes green.... end of a long road.

It's pretty great to have incremental compilation working!

My next steps I think are:

1. documenting using bosatsu to generate json
2. try to write more libraries
3. use graal native-image to generate the bazel rule so the we don't pay the 1.5s JVM startup and don't need to worry about workers
4. add optimization/normalization rules that transform TypedExpr to TypedExpr.
5. add more tests of coercion in TypedExpr. It's a weaker kind of type inference, but I don't think it is solved: you know a type and cast it down to a narrower type.
6. scala API for embedding in scala programs directly without having to go through JSON -> circe, or whatever.

[johnynek]

The coverage issue is due to Main not having coverage in tests, but is tested by the bosatsu bazel rules.