[FIRRTL][IMDCE] missed optimization to delete dead instance

Nobody has claimed this yet.

I think that IMDCE is not deleting instances of public modules, which should be allowed. Given the following test case:

```firrtl
FIRRTL version 4.0.0
circuit Foo:
public module Bar:

public module Foo:
inst bar of Bar
```
Which generates the following IR:
```mlir
firrtl.circuit "Foo" {
firrtl.module @Bar() attributes {convention = #firrtl} {
}
firrtl.module @Foo() attributes {convention = #firrtl} {
firrtl.instance bar @Bar()
}
}
```
IMDCE results in the following error message and IR (which is identical to the input):
```
imdcething.fir:3:10: warning: module `Bar` is empty but cannot be removed because the module is public
public module Bar:
^
```
```mlir
// -----// IR Dump After IMDeadCodeElim (firrtl-imdeadcodeelim) //----- //
module {
firrtl.circuit "Foo" {
firrtl.module @Bar() attributes {convention = #firrtl} {
}
firrtl.module @Foo() attributes {convention = #firrtl} {
firrtl.instance bar @Bar()
}
}
}
```
Which overall results in the following verilog:
```verilog
// Generated by CIRCT unknown git version
module Bar();
endmodule

module Foo();
Bar bar ();
endmodule
```

There is no reason for the instance of `bar of Bar` to have stayed alive, and I expect that it should have been deleted.

Start with the FIRRTL IMDCE pass and reproduce the supplied Foo/Bar example, checking why the public Bar instance remains alive. Done means IMDCE removes the unused instance of Bar while preserving the public module and produces Verilog without the instantiation.

Written by the indexing model from the issue text.