fix: fix a couple issues with duplicate entries and specifier (submodule) matching

[trentm]

This clarifies the hook matching section (as discussed a bit at #187 (comment)) and fixes a couple issues when the Hook arg has duplicate entries or matches a module and a sub-module (and internals: true is being used).

Prerequisites

• This assumes Revert "feat: Optionally hook internal paths like require-in-the-middle (#194)" #236 gets merged first (i.e. that experimentalPatchInternals has been dropped). I'll deal with merge conflicts if/when that is in.

Details

1. Change specifier (sub-module) match to return name==specifier

Given this hook that matches sub-module imports / entry points:

Hook(['an-esm-module-not-using-exports-field/lib/foo.js', '@langchain/core/vectorstores'], (exports, name, baseDir) => {
  console.log('HIT: name=%s', name)
})
await import('an-esm-module-not-using-exports-field/lib/foo.js')
await import('@langchain/core/vectorstores')

Currently the on-import callback gets these name values:

HIT: name=an-esm-module-not-using-exports-field
HIT: name=@langchain/core

This PR changes the returned name to be the specifier:

HIT: name=an-esm-module-not-using-exports-field/lib/foo.js
HIT: name=@langchain/core/vectorstores

Which strikes me as more useful when using specifier matching. Otherwise a single hook that matches multiple entry-points in the same package couldn't differentiate with the name which was hooked.

2. Fix behaviour of internals: true when there are duplicate module entries

Given:

Hook(['@langchain/core', '@langchain/core'], (exports, name, baseDir) => {
  console.log('HIT: name=%s', name)
})
await import('@langchain/core')

Currently the Hook returns matches for each entry:

HIT: name=@langchain/core
HIT: name=@langchain/core

However, with internals:true:

Hook([
  '@langchain/core',
  '@langchain/core'
], { internals: true }, (exports, name, baseDir) => {
  console.log('HIT: name=%s', name)
})
await import('@langchain/core')

This get messed up (because name var handling in the match loop is keeping state,
assuming there will only ever be one match).

HIT: name=@langchain/core/dist/index.js
HIT: name=@langchain/core/dist/index.js/dist/index.js  <-- This is wrong.

The same issue also leads to bogus name values when a module and an entry-point in that module are being matched:

Hook([
  '@langchain/core',
  '@langchain/core/vectorstores'
], { internals: true }, (exports, name, baseDir) => {
  console.log('HIT: name=%s', name)
})
await import('@langchain/core/vectorstores')

Results in:

...
HIT: name=@langchain/core/dist/utils/zod-to-json-schema/index.js
HIT: name=@langchain/core/dist/utils/json_schema.js
HIT: name=@langchain/core/dist/runnables/graph.js
HIT: name=@langchain/core/dist/runnables/wrappers.js
HIT: name=@langchain/core/dist/runnables/iter.js
HIT: name=@langchain/core/dist/runnables/base.js
HIT: name=@langchain/core/dist/retrievers/index.js
HIT: name=@langchain/core/dist/vectorstores.js
HIT: name=@langchain/core/dist/vectorstores.js/dist/vectorstores.js   <--- This one is wrong

This PR fixes both those cases (the part using internalPath instead of mutating name).

[BridgeAR]

I think matching this with RITM would be good. I would be fine to do that as a follow-up PR.

[trentm]

@BridgeAR Please see #239 which discusses this. I don't know how to fix this difference very easily.