Ensure every constraint to wild has its own PSL and remove the `AtomSourceMap`?

[mattmccutchen-cci]

Currently, when 3C adds a constraint (including a constraint to wild), it can specify a PSL, which is supposed to identify the code that resulted in the constraint between the two atoms. When 3C reports a root cause of wildness, it uses the PSL of the constraint to wild if one is specified, otherwise it falls back to the AtomSourceMap, which is supposed to give the PSL of the definition of the program element that was constrained to wild. This fallback has pros and cons, though they each have caveats:

• Pro: In many cases (e.g., an element defined in unwritable code), we don't have a PSL other than the element definition anyway, so relying on the AtomSourceMap could save us some duplicate code at call sites that add constraints. However, using the AtomSourceMap may not be the best way to solve this problem: we might prefer to factor out the logic that determines the PSL for the element definition and use it at each call site. Then we could make the PSL parameter required to force ourselves to think about what PSL makes sense in each case.
• Con: The automatic fallback can make bugs in the determination of the PSL associated with the constraint more confusing to track down. However, this could be addressed by adding more debug output to 3C that is conditional on a flag, which is probably a useful thing to have in general.

Mike said at yesterday's meeting that as we increasingly prioritize user-friendliess of 3C, we should commit to specifying a good PSL for every constraint to wildness and drop the AtomSourceMap fallback. I'm not completely convinced, but I'm going ahead and filing an issue for it.

[kyleheadley]

I ran into the AtomSourceMap when improving conflicting constrints in #708. Conflicts happen at an atom, and while that PR uses the neighboring constraints with their PSLs, it also references the atom itself, which doesn't have a PSL(it uses the default). This is improved in code that runs later by reading the AtomSourceMap.

So the map stores the PSLs for declarations, which can be just as useful for user-friendliness as the constraint generating locations.