[#271] Parametrize initDebugger to take GHC session rather than use HieBios

[Saizan]

There are currently two issues:

1. Understand the current interaction with HIE
2. Figure out how much of the current runDebugger we can/want/should to delegate to ghc-worker.

@alt-romes I pushed a couple of questions as FIXMEs if you can take a look.

Also, RE 2, since ghc-worker is designed to compile one module at a time it has code that takes care of building the right home unit env.

This could mean we can drop a bunch of what's happening in runDebugger but then the code paths between using HIE and ghc-worker would differ significantly. If we think that'd become a maintenance headache we can also just take advantage of any EPS, ModuleGraph, ... already built, while keeping our own code to build the UnitEnv from Targets and DynFlags.

On the other hand it would probably scale better to let buck/the worker handle building the necessary modules? (Even if (at least some of) the modules need to be rebuilt with InsertBreakpoint and other debug specific flags.)

Either way I'm also trying to better understand what's going on in the UnitEnv/HscEnv, I'm thinking I need to serialize it to JSON in all its gory details to wrap my head around it.

[alt-romes]

The ubuntu runner was hanging.

I've made some changes to the error/request abortion cycle which should make the DAP testsuite better at exiting with an error rather than hanging in #260 (notably, 5a163db)