[FIRRTL] LowerLayers: Inline enabled layers

[rwy7]

• move layer utilities to LayerSet.h header
  • adds new source LayerSet.cpp which contains the implementations of the layer utilities
• when lowering a layer block, if the layer is enabled, inline it instead, by splicing the layer block's body into the parent block.

[seldridge]

Nice. This will improve the output quality a fair amount by avoiding unnecessary layer cruft.

Two things:

1. Can the PR include an end-to-end test to lock-in the Verilog quality improvement?
2. Is this still ABI-preserving? This would be obvious with (1). I.e., what happens if there is a public module that has an enabled layer, yet it runs into this optimization?

I did not review that no changes were made in the factoring out of the the layer utilities into the new header/file. While not required, that can be factored out into a separate commit / rebase-and-merged. Not required.

[seldridge]

This is cleaner without the .template .

Suggested change

	.Case<FModuleOp>([&](auto op) {
	LayerSet enabledLayers;
	enabledLayers.insert_range(
	op.getLayersAttr().template getAsRange<SymbolRefAttr>());
	.Case<FModuleOp>([&](FModuleOp op) {
	LayerSet enabledLayers;
	enabledLayers.insert_range(
	op.getLayersAttr().getAsRange<SymbolRefAttr>());

[seldridge]

Super nit: if this is a frequent pattern, i.e., "get a layer set of a module", then this would be good as an alternative constructor for LayerSet. I think this would require making LayerSet an actual class as opposed to a type alias.

[seldridge]

Nice splice! It's clean and super efficient. 💯