Bridge/LSP hardening: shared timeout constant, RFC 3986 encoding test, retry-loop test coverage

[bug-ops]

Summary

Grouped triage batch of three low-priority (P3) issues, all touching LSP request-handling internals in mcpls-core (bridge/lsp modules), no cross-dependencies:

• bridge: extract shared LSP request timeout constant instead of repeating Duration::from_secs(30) #231 — extract DEFAULT_LSP_TIMEOUT/COMPLETIONS_LSP_TIMEOUT constants in bridge::translator, replacing 17 duplicated Duration::from_secs(...) literals. Also corrects timeout_seconds documentation in configuration.md/troubleshooting.md, which incorrectly implied it bounds per-request LSP timeouts — it only bounds the initialize handshake.
• encode_rfc3986_path_chars: incomplete character set (missing { } and backtick) #168 — adds a regression test pinning RFC 3986 §2.2 percent-encoding behavior of encode_rfc3986_path_chars/try_path_to_uri. Confirms the underlying bug was already fixed as a side effect of fix(core): encode reserved path chars in file URIs #151 (the url crate already percent-encodes backtick, {, } on serialization); no production code change needed for those three characters.
• test(bridge): add unit tests for should_retrigger() and retry exhaustion path #161 — adds unit test coverage for LspClient::should_retrigger() and its wiring into the ServerCancelled (-32802) retry loop: gate-function cases, full retry exhaustion, immediate short-circuit on retriggerRequest: false, and the retry-success happy path.

Out of scope (follow-up issues to be filed)

• bridge::resources::make_uri builds lsp-diagnostics:// URIs without going through encode_rfc3986_path_chars, so [ ] ^ | can still leak unencoded on that path. Outside encode_rfc3986_path_chars: incomplete character set (missing { } and backtick) #168's literal scope (which names only encode_rfc3986_path_chars); CHANGELOG wording is scoped accordingly.
• Near-identical fake LSP transport test helpers now exist in three files (state.rs, translator.rs, client.rs); a shared test-utils module would reduce duplication but is out of scope for this batch.

Test plan

• cargo +nightly fmt --all -- --check
• cargo clippy --all-targets --all-features --workspace -- -D warnings
• cargo nextest run --workspace --all-features --lib --bins (513 passed, 0 failed)
• RUSTDOCFLAGS="-D warnings" cargo doc --no-deps --all-features --workspace

[Copilot]

Pull request overview

This PR hardens mcpls’s MCP↔LSP bridge by (1) centralizing the default per-request LSP timeouts, (2) adding a regression test for RFC 3986 reserved-character percent-encoding in file:// URI conversion, and (3) adding unit tests that pin ServerCancelled (-32802) retry-loop behavior in the LSP client.

Changes:

• Extract shared DEFAULT_LSP_TIMEOUT / COMPLETIONS_LSP_TIMEOUT constants in bridge::translator and update docs to clarify timeout_seconds only applies to the initialize handshake.
• Add regression coverage ensuring all RFC 3986 §2.2 “other reserved” characters are percent-encoded in try_path_to_uri / encode_rfc3986_path_chars.
• Add tests for LspClient::should_retrigger() and end-to-end retry-loop outcomes (exhaustion, short-circuit on retriggerRequest: false, and retry-success).

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
docs/user-guide/troubleshooting.md	Clarifies what timeout_seconds affects and gives better guidance for large-project initialization vs post-init timeouts.
docs/user-guide/configuration.md	Corrects timeout_seconds documentation to “initialize-only” and distinguishes it from fixed per-request timeouts.
crates/mcpls-core/src/lsp/client.rs	Adds should_retrigger() and retry-loop unit tests using a fake stdio transport.
crates/mcpls-core/src/bridge/translator.rs	Introduces shared timeout constants and replaces repeated Duration::from_secs(...) literals.
crates/mcpls-core/src/bridge/state.rs	Adds RFC 3986 percent-encoding regression test and expands/clarifies encoding documentation.
CHANGELOG.md	Documents the timeout constant extraction and the new regression/unit test coverage.