[ExtractTestCode] Specify non-empty unqiue port names

[prithayan]

Update ETC to add non-empty and unique port names. The name itself is not important, but this exposes a bug in inconsistent port names with the hw.instance and the ModuleType.
This is a temporary fix, to unblock downstream tools, while the actual bug is fixed.

[uenoku]

Maybe can we do something simpler and more stable naming like this?

Suggested change

	b.getStringAttr(getNameForPort(port.value(), srcPorts).getValue() +
	auto name = getNameForPort(port.value(), srcPorts);
	name = name.getValue().empty()? b.getStringAttr("arg" + Twine(port.index()));

[uenoku]

Also “arg” seems to be a bit weird to me, can we do name this “port”?

[prithayan]

So, the ETC introduces port name conflicts, that exposes the same bug, inconsistent port names on instance and module. So, we either check and rename only conflicting port names or create unique names by suffixing with the port index. I assumed the port names donot matter here, so suffixed with the index for all ports, instead of adding a check.
I am working on another PR to fix the LegalizeNames to update argNames on the instance op also. So, we could skip the unique names logic here.

[prithayan]

After looking at LegalizeNames realized the conflicting port names case cannot be handled there. This is inherently a problem of SSA names leaking to the module type and port names.
Fixed the logic here to create unique names only on name conflict.

[jackkoenig]

LGTM but may want someone who knows the code better than me to review

[uenoku]

Can we use Namespace?

[prithayan]

Right, we should use the Namespace. Also added a check for empty string to use a stable and deterministic naming for them.
For such use cases, we need a Namespace::newName which adds a given suffix only on conflict, the current version always adds the suffix.

[uenoku]

LGTM