Filing a set of futures relating to module init order / main modules

[bradcray]

These tests threw me for a loop for awhile this evening (along the line
of "Why is my code not running?!?!"

The challenges relate to which modules are candidates for the main
module and which modules' init functions (top-level code) are run.
What we have today is rationalizable, but here I propose that some
warnings might help avoid confusion in cases like this.

I think what we have today is:

• modules that are 'require'd or named on the command-line are candidates
  for main modules; if a module is not in this group, it is not
  a candidate for the main module and its main() will not be run.
• among the modules that are candidates for main modules, if only one has
  main(), it is the main module
• once the main module is determined, module initialization is determined
  by walking the tree of 'use's starting from the 'main' module
• if a module is named on the command-line but is not reached via this
  tree, its initialization (top-level code) will not be run.

As I say in the .future, I think this is rationalizable, but it can be confusing
in some cases. To that end, I propose putting a warning when modules
fall into bullet 4's case to avoid surprises.