fix[tool]: add missing internal functions to metadata

[sandbubbles]

What I did

Add internal functions from imported modules into metadata output. Issue #4312 but reverse the imports.

How I did it

Print all reachable functions rather than just functions that are present in the module.

How to verify it

Commit message

this commit adds all reachable internal functions to the `-f metadata`
output. this was overlooked during the 0.4.0 release - there could
be internal functions in submodules which didn't get added to the `-f
metadata` output, since the routine to build the `-f metadata` output
only looked at the top-level module.

this commit also makes a change to the format of the `.function_info`
keys - since function names are not guaranteed to be globally unique
anymore (that is, two different modules may both contain a function
named `foo`), this commit adds a disambiguating function id to each
key in `.function_info`. (it uses the function_id computed during IR
generation since it is guaranteed to be unique). the function name
is still present in the function metadata under the `.name` key, but
downstream tooling may have to adapt if they had the assumption that
the keys in `.function_info` were parseable names. for illustration,
here is an example of function metadata prior to and after this commit.

prior to this commit:
```json
{
  "function_info": {
    // valid identifier, but not necessarily unique
    "foo": {
      "name": "foo",
      "_ir_identifier": "internal 0 foo()",
      ...
    }
  }
}
```

after this commit:
```json
{
  "function_info": {
    // not a valid identifier, but is unique. for the function name, see
    // the "name" field below.
    "foo (0)": {
      "name": "foo",
      "_ir_identifier": "internal 0 foo()",
      ...
    }
  }
}
```

Description for the changelog

Cute Animal Picture

[charles-cooper]

what does inf_t stand for?

[sandbubbles]

internal function and the t cause its the type - ill rename it to something clearer

[charles-cooper]

hmm noticing something messed up -- what if we have two functions with the same name in different modules?

[charles-cooper]

i think there can actually be collisions between path since it's not resolved (e.g. depending on which search path is used)

[charles-cooper]

i think we need to find a way to export the internal functions which doesn't have any collisions in the dict. i.e. it is injective from reachable functions to functions which actually appear in the output.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.74%. Comparing base (f38b61a) to head (4a95adb).
Report is 8 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #4328      +/-   ##
==========================================
- Coverage   90.99%   88.74%   -2.25%     
==========================================
  Files         112      112              
  Lines       16017    16032      +15     
  Branches     2696     2698       +2     
==========================================
- Hits        14574    14227     -347     
- Misses       1005     1289     +284     
- Partials      438      516      +78     

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

[sandbubbles]

i think we need to find a way to export the internal functions which doesn't have any collisions in the dict. i.e. it is injective from reachable functions to functions which actually appear in the output.

changed it to use _metadata["type"]._id instead of path

[charles-cooper]

i think we need to find a way to export the internal functions which doesn't have any collisions in the dict. i.e. it is injective from reachable functions to functions which actually appear in the output.

changed it to use _metadata["type"]._id instead of path

sorry, ModuleT._id isn't guaranteed to be unique because it uses .path instead of .resolved_path 😬

[sandbubbles]

We are now using the IR function id (which is unique). This should be good for another review.

[charles-cooper]

this lgtm, i will check with downstream users to see if they can use the new function identifier format