Improve compilation time

[djspiewak]

This is a decent exemplar: https://github.com/slamdata/quasar/pull/4127#discussion_r282934462

Shims imposes some noticeable compile-time overhead during implicit search. In some scopes, this can result in some very dramatic differences. Considering that the whole implicit search process is effectively controlled by the custom macro, it should be possible for us to do better on this front.

[djspiewak]

Early returns are not promising:

2.1.0

sbt:quasar-qsu> compile
[info] Compiling 1 Scala source to /Users/daniel/Development/slamdata/quasar/qsu/target/scala-2.12/classes ...
[success] Total time: 19 s, completed Oct 23, 2019 6:27:56 PM
sbt:quasar-qsu> compile
[info] Compiling 1 Scala source to /Users/daniel/Development/slamdata/quasar/qsu/target/scala-2.12/classes ...
[success] Total time: 1 s, completed Oct 23, 2019 6:28:02 PM


2.1-1aafef6

sbt:quasar-qsu> compile
[info] Compiling 1 Scala source to /Users/daniel/Development/slamdata/quasar/qsu/target/scala-2.12/classes ...
[success] Total time: 30 s, completed Oct 23, 2019 6:33:47 PM
sbt:quasar-qsu> compile
[info] Compiling 1 Scala source to /Users/daniel/Development/slamdata/quasar/qsu/target/scala-2.12/classes ...
[success] Total time: 2 s, completed Oct 23, 2019 6:33:53 PM