Remove `drop` ops using linearizer

[mark-koch]

See #946

[acl-cqc]

Some of this will have to be done in Hugr, but I see two ways:

• Extend replace_parametrized_op so that the callback is not Fn(&[TypeArg]) -> .... but Fn(&[TypeArg], &ReplaceTypes) -> .... or Fn(&[TypeArg], &DelegatingLinearizer) -> ..... This would allow registering a callback for drop that returns linearizer.copy_discard_op(type_args[0], 0).
  • There is probably some hacky way to duplicate the knowledge of what-all-the-affine-types-are (and their drop operations), currently stored in the (Delegating)Linearizer, into that callback, but I'll dismiss that for now.
• Extend ReplaceTypes to look not just at the (output) signature of the replacement operation, but also any internal ports if it's a subtree: https://github.com/CQCL/hugr/blob/939912a83be81a33e9619ccd720ea665886be978/hugr-passes/src/replace_types.rs#L533-L547

[acl-cqc]

Closed by #979