feat!: Extend LowerTypes pass to linearize by inserting copy/discard

[acl-cqc]

• Does not handle nonlocal edges. This is much more involved, not part of this PR. I note feat(hugr-passes)!: Add UnNonLocalPass #1912
• handlers for arrays follow in feat: ReplaceTypes: handlers for array constants + linearization #2023

BREAKING CHANGE: OpReplacement renamed to NodeTemplate. Note this is not a breaking change if this PR goes in the same release as #1989 (which introduced OpReplacement)

[hugrbot]

This PR contains breaking changes to the public Rust API.

cargo-semver-checks summary

--- failure enum_missing: pub enum removed or renamed ---

Description:
A publicly-visible enum cannot be imported by its prior path. A `pub use` may have been removed, or the enum itself may have been renamed or removed entirely.
      ref: https://doc.rust-lang.org/cargo/reference/semver.html#item-remove
     impl: https://github.com/obi1kenobi/cargo-semver-checks/tree/v0.40.0/src/lints/enum_missing.ron

Failed in:
enum hugr_passes::replace_types::OpReplacement, previously in file /home/runner/work/hugr/hugr/BASELINE_BRANCH/hugr-passes/src/replace_types.rs:28

[acl-cqc]

@doug-q ok so with the OpReplacement -> NodeTemplate rename the pressure of time kicks in, but I think I've done most of the things you suggest (good comments, thank you!) - the only thing that remains is the errors: I kinda disagree with your point about builder errors, I think those are our mistakes, but I've noted in my replies a couple of things I would be happy to make into reported errors (easy ones, too!), what do you think?

EDIT: I've also remove the linearize_x methods from ReplaceTypes and just expose the Linearizer directly. This was better in terms of giving one place in which to concentrate docs; and in the future, one can imagine giving the linearizer a run method that, without replacing types, just checks every outport of every node. (I think this can come after #1895, given that validation before Linearizer will fail! But we can keep the idea that ReplaceTypes has its own Linearizer that it uses only on nodes that it has changed.)

[acl-cqc]

So perhaps that this one takes a callback, supporting arbitrary N, whereas register has only 0-ary discard and 2-ary copy, might be the bigger difference; both can be used for monomorphic types. It might be better to rename, thus

• this one to register_callback or register_with_callback or even just register?
• register maybe to register_binary / register_2way / register_(dup/copy)_discard / register_simple ??

[doug-q]

I like register_simple/register_callback best, but do not feel strongly about it..

[acl-cqc]

SGTM, done

[acl-cqc]

Ok @doug-q fire away. I tried using &dyn Linearizer rather than the rather-odd-seeming-on-its-own CallbackHandler....yes dyn is inefficient but that seemed a lot cleaner...however it required making Linearizer object-safe and other changes (but you can see cc4fd68)

[doug-q]

Looks great, thanks Alan. I will fix CI.