Support LSP 3.17 positionEncoding negotiation

[ahogappa]

Summary

Implements LSP 3.17 positionEncoding negotiation. Today TypeProf hardcodes UTF-16LE columns and omits positionEncoding from its capabilities response, so modern clients that prefer UTF-8 (Helix, Zed, Neovim) all fall back to UTF-16.

This PR also exposes the encoding as Service.new(position_encoding:) for embedders.
The motivating use case is ruby-minify, which cross-references TypeProf nodes against Prism re-parses to drive variable/method renaming.
Prism's native columns are byte-based — equivalent to LSP utf-8 — so Service.new(position_encoding: Encoding::UTF_8) makes the two coordinate systems line up.
Without this option, ruby-minify reaches into TypeProf via node.instance_variable_get(:@raw_node).location to bypass UTF-16 columns.

Behavior change

None by default. Clients that don't send general.positionEncodings, and embedders that don't pass position_encoding, still get UTF-16LE.

Implementation note

The server picks the first mutually-supported encoding from the client's preference-ordered list (utf-8 / utf-16 / utf-32), falling back to utf-16 per spec.

FileContext#column_offsets_for returns columns in the configured encoding. For UTF-8 it uses Prism's native start_column / end_column (= byte offsets) — note that Prism's code_units_cache(Encoding::UTF_8) reports code points rather than bytes, which would violate the LSP spec.

Verification

End-to-end against Helix, which proposes general.positionEncodings: ["utf-8", "utf-32", "utf-16"]. Source — line 4 references undefined あfoo, where あ is 3 bytes in UTF-8 / 1 code unit in UTF-16:

def foo(n)
  n
end

あfoo(1, 2)

Helix helix_lsp::transport log — InitializeResult

Master (ruby/typeprof 0.31.1):

typeprof <- {"id":0,"result":{"capabilities":{"textDocumentSync":{...},"hoverProvider":true,...,"referencesProvider":true},"serverInfo":{"name":"typeprof","version":"0.31.1"}},"jsonrpc":"2.0"}

This PR:

typeprof <- {"id":0,"result":{"capabilities":{"positionEncoding":"utf-8","textDocumentSync":{...},"hoverProvider":true,...,"referencesProvider":true},"serverInfo":{"name":"typeprof","version":"0.31.1"}},"jsonrpc":"2.0"}

The only difference is the new "positionEncoding":"utf-8" field, which signals that subsequent column values use UTF-8 byte offsets.

• Unit tests in test/core/position_encoding_test.rb (UTF-8 / UTF-16 / UTF-32 columns).
• LSP tests in test/lsp/lsp_test.rb (negotiation matrix).
• End-to-end against Helix: confirmed positionEncoding: "utf-8" response and correct diagnostic range on a non-ASCII identifier.

Spec: https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#textDocuments

[ahogappa]

FYI: given that Ruby's default source encoding is UTF-8, UTF-8 would arguably be a more natural default.
However, switching from UTF-16LE would be a breaking change for existing users, so this PR keeps UTF-16LE.

[mame]

Thanks, I would like to give it a try!