F# first-class parity: document symbols, signature help, hierarchy routing fix (+95% sidecar coverage)

[MelbourneDeveloper]

Summary

Brings F# to first-class parity for the remaining LSP outline/intelligence features and fixes a routing bug that silently broke F# call/type hierarchy follow-up requests. Also lands the F# sidecar coverage push (→95%) and the VS Code extension unit-test wave that were developed on this branch.

New F# features

• [FS-DOCSYMBOL] textDocument/documentSymbol — F# document symbols via FCS GetNavigationItems (parse-only, so it responds without a cracked project). The host routes .fs to the F# sidecar; .cs stays on tree-sitter. (document_symbols.rs, FSharpSymbols.fs)
• [FS-SIGHELP] textDocument/signatureHelp — overload signatures via FCS GetMethods; capability now advertised with (/, triggers. (signature_help.rs, FSharpSignature.fs)

Bug fix — F# hierarchy routing

callHierarchy/incomingCalls/outgoingCalls and typeHierarchy/super/subtypes carry the document URI inside params.item, not params.textDocument. extract_document_uri only checked textDocument/data, so these follow-up requests defaulted to the C# sidecar for .fs files — which is why F# incoming calls came back empty. Now it also reads params.item.uri. (main.rs)

Tests & coverage

• New sidecar IPC + direct unit tests for document symbols and signature help, plus Rust unit tests for the new handlers.
• F# sidecar coverage 94.77% (threshold 95.00, within tolerance); sharplsp 92.12%; all per-package floors held.
• The four previously-aspirational F# e2e suites (document symbols ×2, signature help, call-hierarchy incoming) now pass against the real release-built LSP + FCS sidecar.

Validation (local, matches CI)

• make _lint-rust / _lint-zed / _lint-dotnet / csharpier / prettier / eslint+tsc — green
• make _test-rust — 611 tests, coverage green
• make _test-dotnet — 660 tests (C# 316, F# 301, common 43), coverage green
• make _test-vsix — 1295 passing; the four target F# tests fixed
• Deslop: no new duplication clusters

Note: one pre-existing, order-dependent C# folding screenshot test (editor.foldAll timing) can flake locally in isolation; it is unrelated to this change (the baseline fails it identically in isolation) and passes in the full suite / on CI.