[query] support emit-level parameters

[cseed]

It might not look like it, but I promise, I'm trying to keep these as small as possible.

Summary of changes:

• EmitMethodBuilder, etc. no longer implement WrappedMethodBuilder, etc. but they do proxy the necessary functions.
• Emit method params and return values are now statically either: an Code[T] with type TypeInfo[T], or an EmitCode, with type PType.
• Added ParamType to wrap the two parameter type options, and Param to wrap the two code options, with implicit conversions.
• A bunch of methods now have emit and code variants: getEmitParam, getCodeParam, invokeCode, invokeEmit (where code, emit refer to the return value).
• I made the minimal changes to get things working again. Some code still code parameters to pass emit values. I will fix these in a later PR.
• Where possible, make the class implemented by Compile concrete rather than generic. I think this can be pushed through the entire code base and we can remove the option to build generic methods. We should have done this a long time ago.
• ModuleBuilder can now create type-specialized tuple types. This is used for EmitCode return values. I'm not sure if this is actually used yet.
• Require RVDType rowType to be required. Require TypeValue global type to be required. Fix lots of places to make this true. In a few spots (e.g. TableMap{Rows, Globals}), I had to wrap the IR being compiled in a Coalesce with a Die to make sure the return type is required.
• Cleaned up the dependent function interface to be closer to what we have now with MethodBuilder, etc. DependentFunctionBuilder is now just an apply_method: DependentMethodBuilder, EmitFunctionBuilder analogously. DependentMethodBuilder wraps a MethodBuilder, EmitMethodBuilder wraps a DependentMethodBuilder and an EmitMethodBuilder.
• Add equality comparison to TypeInfo[_]
• Add methods to convert IndexedSeq[Code[_]] to/from PCode and EmitCode. These are used to pass EmitCode as arguments to method invocation. If an emit parameter is required, the missingness boolean is omitted, otherwise it is present.

Furthermore, this change also adds requiredness to many things and improves ptype interfaces:

• added PType.literalPType that infers PTypes from Scala literals, use in a few places (emit for Literal, BroadcastRegionValue constructor from annotation, etc.)
• require Table global and row types to be required
• same for MatrixValue, but also cols and entries (the entries array, not individual entries, which an be missing)
• Don't upcast globals in TableKeyBy and TableOrderBy
• added EType setRequired
• AbstractCodecSpecs assert row and global etypes are present at the toplevel, and setRequired(true) if they are coming from encoders written by previous versions
• rename PType.copyFromType to PType.copyFromAdddres. Modify it so it can "downcast": convert to a PType with greater requiredness. This is used in converting TableValues to MatrixValues to satisfy the requiredness assertions.

[cseed]

The rebase was a little messy. I got it compiling, but there will probably be some minor issues with the tests. It was passing before, so I expect the overall code to be sound.

[cseed]

Looks like some Python failures. It will take me a little while to track them down, but they all look minor and of two forms I understand:

• RVDType being constructed with the row type being optional,
• and incorrect type signatures when building emit methods due to mismatched missingness.
  Go ahead and review while I fix them.

[chrisvittal]

To my eye, this was super clean. A few nits. Substantive parts of the change all look great.

[cseed]

A miracle. It finally passed. That was a real slog. I pushed a bunch of non-trivial changes, so it is probably good if you give a skeptical, fresh look. Summary of new changes:

• added PType.literalPType that infers PTypes from Scala literals, use in a few places (emit for Literal, BroadcastRegionValue constructor from annotation, etc.)
• require Table global and row types to be required
• same for MatrixValue, but also cols and entries (the entries array, not individual entries, which an be missing)
• Don't upcast globals in TableKeyBy and TableOrderBy
• added EType setRequired
• AbstractCodecSpecs assert row and global etypes are present at the toplevel, and setRequired(true) if they are coming from encoders written by previous versions
• rename PType.copyFromType to PType.copyFromAdddres. Modify it so it can "downcast": convert to a PType with greater requiredness. This is used in converting TableValues to MatrixValues to satisfy the requiredness assertions.

Let me know if you have any questions.

addressed comments, passing

[chrisvittal]

I've been following along. So I've seen a lot of the new stuff already. I'll take another look.

[chrisvittal]

Let's see what shakes out.