feat(lsp): add per-request LSP timeout configuration

[bug-ops]

Summary

• LspServerConfig::timeout_seconds only ever bounded the initialize handshake; every per-request LSP round trip (hover, definition, references, etc.) used fixed DEFAULT_LSP_TIMEOUT/COMPLETIONS_LSP_TIMEOUT constants that users could not tune, despite pre-Bridge/LSP hardening: shared timeout constant, RFC 3986 encoding test, retry-loop test coverage #264 documentation implying otherwise.
• Adds LspServerConfig::request_timeout_seconds (default 30, additive field) as a distinct per-request knob, reached via new LspClient::request_timeout()/completion_timeout() accessors — no Translator signature changes required, all 17 call sites in bridge::translator swapped from the fixed constants to the accessors.
• completion_timeout() clamps to at most 10s regardless of the configured value (an explicit MVP decision, not an oversight — completions are the most latency-sensitive tool and the MCP client's own timeout leaves no margin for a longer wait).
• ServerConfig::validate() now rejects timeout_seconds == 0 and request_timeout_seconds == 0; request_timeout() also clamps to a minimum of 1s to cover the programmatic serve() entry point, which does not go through validate(). The same .max(1) guard was also applied to the pre-existing handshake timeout_seconds on the initialize call in lifecycle.rs, closing the identical gap there.
• Documents the real worst-case latency for one tool call: 4 * request_timeout_seconds + 3.5s, since LspClient::request retries up to 4 attempts on a -32802 response.
• Updates docs/user-guide/configuration.md, docs/user-guide/troubleshooting.md, and docs/user-guide/tools-reference.md to remove the stale "not configurable" claim and document the new field.

Process

Went through the full team-develop pipeline: architect plan → adversarial critic (2 rounds, first returned significant on the completion-timeout formula, retry-ceiling math, and a missing zero-value validation gap; all three resolved and the revision approved as minor) → implementation → parallel validation (tester, perf, security, and an implementation-level adversarial critic — the latter caught one blocking rustdoc inaccuracy plus the timeout_seconds .max(1) gap noted above, both fixed) → final code review, approved. Rebased onto main after merge of #268-#271 (conflict only in CHANGELOG.md; one rebase-introduced test fixture needed the new field added).

One low-severity, non-blocking item was surfaced during security review and is left for a follow-up issue (#273) rather than blocking this PR:

• request_timeout_seconds has no upper bound — an extremely large configured value silently disables the timeout via tokio's ~30yr fallback deadline.

Test plan

• cargo +nightly fmt --all -- --check
• cargo clippy --all-targets --all-features --workspace -- -D warnings
• cargo nextest run --workspace --all-features --lib --bins (574 passed)
• cargo test --doc --all-features (9 passed, including 2 new doctests)
• RUSTDOCFLAGS="-D warnings" cargo doc --no-deps --all-features
• New tests cover: default identity (30s/10s reproduces prior hardcoded behavior), completion-timeout clamp table, zero-value clamp on both config paths, per-server independence, TOML backward compatibility (timeout_seconds present, request_timeout_seconds absent), serde roundtrip

Closes #267

[Copilot]

Pull request overview

This PR adds a per-request, per-LSP-server timeout configuration knob to mcpls, aligning runtime behavior with user expectations and prior documentation by making tool-call LSP round-trip timeouts configurable instead of hardcoded.

Changes:

• Introduces LspServerConfig::request_timeout_seconds (default 30s) and threads it through LspClient::request_timeout() / completion_timeout() so bridge::translator uses configured timeouts instead of fixed constants.
• Enforces non-zero timeouts in TOML-loaded configs (ServerConfig::validate() rejects timeout_seconds == 0 and request_timeout_seconds == 0) and clamps programmatic configs to at least 1s on the runtime path.
• Updates documentation + changelog to describe the new field, the completion timeout cap, and the retry-based worst-case latency math.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
README.md	Adds request_timeout_seconds to the example LSP server config snippet.
docs/user-guide/troubleshooting.md	Updates troubleshooting guidance to use request_timeout_seconds and documents worst-case latency including retries.
docs/user-guide/tools-reference.md	Fixes timeout remediation guidance to reference request_timeout_seconds.
docs/user-guide/configuration.md	Documents request_timeout_seconds, including retry ceiling math and the completion timeout cap.
crates/mcpls-core/tests/integration/rust_analyzer_tests.rs	Updates integration test config structs to include the new field.
crates/mcpls-core/src/lsp/lifecycle.rs	Applies a .max(1) clamp to timeout_seconds for initialize on the non-validated serve() path.
crates/mcpls-core/src/lsp/client.rs	Adds request_timeout() / completion_timeout() accessors, completion cap constant, and unit tests for timeout behavior.
crates/mcpls-core/src/lib.rs	Updates internal test config literals to include request_timeout_seconds.
crates/mcpls-core/src/config/server.rs	Adds the request_timeout_seconds field with serde default + updates canned configs and tests.
crates/mcpls-core/src/config/routing.rs	Updates routing-related tests/config literals for the new field.
crates/mcpls-core/src/config/mod.rs	Validates both timeout fields are non-zero and adds TOML backward-compat/defaulting + validation tests.
crates/mcpls-core/src/bridge/translator.rs	Replaces hardcoded per-request timeouts with client.request_timeout() / client.completion_timeout() across tool handlers.
CHANGELOG.md	Documents the new config field and notes the removal of fixed translator timeout constants.