Refactor the graph functionality to remove the custom module name parser

[f-f]

This refactoring PR orchestrates a few things:

• we remove the custom module name parser that was introduced in Do not error when there are no tests #489, using the new functionality from purs graph instead
• we now call purs graph when instantiating a BuildEnv, so that we get the graph only once at the beginning, and have it handy anytime we are building
• the init command has its own module (we should eventually do this with all the commands)

[colinwahl]

Look great @f-f, just saw one minor improvement.

[colinwahl]

the init command has its own module (we should eventually do this with all the commands)

Do you also intend for all of the commands in Spago.Build to be pulled into their own modules down the line? I could help with moving things around if I know where they should go!

[f-f]

Thank you for the review @colinwahl!

Do you also intend for all of the commands in Spago.Build to be pulled into their own modules down the line? I could help with moving things around if I know where they should go!

That would be great!
So far I created such modules only to break circular dependencies, but I think ideally all the toplevel commands should have their Spago.Command.* module.
Something I'm still not sure about is the granularity of the effort, e.g. if every exported function in the Spago.Build module would go to its own module, or if that would end up being confusing and we should go for bigger modules. In any case the Spago.Build module might not disappear, but remain to hold common things such as runBackend. Or something like that 🙂