feat!: Merge Value into Const

[aborgna-q]

Const was a struct with a Value + a defined type. In practice, the only Value that couldn't report its full type was Sum.

This PR merges the two structs, adding an explicit sum_type to the sum variant.

Also:

• Removes Type::check_type, as this is now unnecessary.
• Splits and clarifies the type check and constant errors;
  SumTypeError is not a ConstTypeError is not a CustomCheckFailure.
  • Adds contexts to those errors, so we get better messages.
• CustomConst is now accessible from ::ops::custom instead of ::values.

This change breaks the serialisation format.

Closes #874.

BREAKING CHANGE: Removed Value, merged into Const

[aborgna-q]

This only returns an Err if tag >= size.
I'm undecided on whether panicking instead, as every caller just unwraps it immediately.

[doug-q]

I prefer using Result over panicking here, because other similar functions (e.g. function above) do. It's a public function so we can't say anything about every caller.

[aborgna-q]

Added an ad-hoc get_const_function_type in lieu of #880, while we decide on that.

[doug-q]

I think the serialisation schema should be changing because the serialisation of const nodes has changed?

We have already bumped the serialisation version to V1 since the last release, so it is fine to change the V1 schema. If we hadn't bumped since the last release this would necessitate bumping.

[aborgna-q]

Yes, here's how the new encoding looks for an (extension) usize:

{
  "v": "Extension",
  "c": {
    "c": "ConstUsize",
    "value": 257
  }
}

And for a [float, usize] + [] const:

{
  "v": "Sum",
  "tag": 0,
  "values": [
    {
      "v": "Extension",
      "c": {
        "c": "ConstUsize",
        "value": 257
      }
    },
    {
      "v": "Extension",
      "c": {
        "c": "CustomSerialized",
        "typ": {
          "t": "Opaque",
          "extension": "arithmetic.float.types",
          "id": "float64",
          "args": [],
          "bound": "C"
        },
        "value": 5.1,
        "extensions": [
          "arithmetic.float.types"
        ]
      }
    }
  ],
  "typ": {
    "s": "General",
    "rows": [
      [
        {
          "t": "I"
        },
        {
          "t": "Opaque",
          "extension": "arithmetic.float.types",
          "id": "float64",
          "args": [],
          "bound": "C"
        }
      ],
      []
    ]
  }
}

[codecov]

Codecov Report

Attention: Patch coverage is 82.48175% with 48 lines in your changes are missing coverage. Please review.

Project coverage is 85.62%. Comparing base (9ef7808) to head (930c5e5).

Files	Patch %	Lines
quantinuum-hugr/src/ops/constant.rs	83.96%	18 Missing and 3 partials ⚠️
quantinuum-hugr/src/ops/constant/custom.rs	65.95%	15 Missing and 1 partial ⚠️
quantinuum-hugr/src/algorithm/const_fold.rs	66.66%	6 Missing ⚠️
...td_extensions/arithmetic/conversions/const_fold.rs	40.00%	2 Missing and 1 partial ⚠️
quantinuum-hugr/src/hugr/rewrite/inline_dfg.rs	0.00%	0 Missing and 1 partial ⚠️
quantinuum-hugr/src/types.rs	80.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #881      +/-   ##
==========================================
+ Coverage   85.49%   85.62%   +0.12%     
==========================================
  Files          78       78              
  Lines       14455    14397      -58     
  Branches    14455    14397      -58     
==========================================
- Hits        12359    12327      -32     
+ Misses       1459     1436      -23     
+ Partials      637      634       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[aborgna-q]

I updated the serialization schema manually (merging the json generated by CQCL/guppylang#172 with the local changes).
Hopefully we can improve this process in the future (see CQCL/guppylang#173).

[doug-q]

Awesome, just a few nits

[doug-q]

I prefer using Result over panicking here, because other similar functions (e.g. function above) do. It's a public function so we can't say anything about every caller.