refactor: use Imenu as symbol completion source

[tninja]

Summary

Refactor @file#symbol completion so Imenu is the canonical symbol-discovery layer, including when the target buffer is using a Tree-sitter major mode.

This keeps the user-facing improvements from #483 (qualified names, source order, header/line annotations) while removing the package-level file-wide Tree-sitter symbol scanner.

Why

#483 treated Imenu and Tree-sitter as alternative symbol providers. In Emacs, Tree-sitter major modes already expose their structural symbol index through Imenu (treesit-simple-imenu / imenu-create-index-function). Bypassing that abstraction makes ai-code-interface.el maintain a second cross-language symbol-discovery implementation and its own function/method/constructor node-type registry.

There is one important nuance: Tree-sitter-backed Imenu indexes are often grouped into flat categories such as Class and Function, so Imenu alone does not always preserve the enclosing-type path needed for Bar.baz.

Changes

• Use imenu--make-index-alist as the single source of which symbols exist and their source positions.
• Preserve semantic Imenu hierarchy when a mode provides it, without treating grouping categories such as Class / Function as name prefixes.
• Restore file-definition order from Imenu entry positions rather than alphabetically sorting candidates.
• For an active Tree-sitter parser, enrich each Imenu-discovered position locally with its enclosing type path, so flat indexes still produce names such as Bar.baz and Outer.Inner.run.
• Use the existing semantic-scope machinery to enrich declaration headers; non-Tree-sitter modes fall back to the source line.
• Remove the file-wide Tree-sitter AST walk and function-node registry introduced by feat(treesit): symbol-level @# completion with qualified names #483.
• Replace provider-specific completion tests with tests for Imenu discovery, flat Tree-sitter Imenu qualification, nested qualification, source ordering, special Imenu entries, and completion annotations.

Design

The abstraction is now:

major mode (Tree-sitter / regexp / LSP / mode-specific) -> Imenu -> ai-code symbol discovery

with Tree-sitter used only as position-local enrichment when an active parser can provide metadata that the Imenu representation omitted.

In particular, this PR does not scan the file-wide Tree-sitter AST to rediscover symbols and does not maintain a new cross-language function/method node-type table.

Validation

• Full ERT suite passes on both push and pull-request workflows.
• Build/melpazoid checks pass on both push and pull-request workflows.
• Codex review's flat Tree-sitter Imenu concern was addressed with regressions for Bar.baz and Outer.Inner.run.
• Branch is squashed to one commit with only the five intended source/test files changed.

[tninja]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8a436a330d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".