Make remote tool-call timeout configurable, lift 60s SDK default

[eshwar-sundar-glean]

Problem

Customer escalation - Chat failure

Every downstream tool call went through callRemoteTool → client.callTool({ name, arguments }) with no options, so the MCP SDK applied its hardcoded DEFAULT_REQUEST_TIMEOUT_MSEC (60s). Any tool that took longer than 60s — invoked via run_tool, find_skills, or the remote passthrough — failed with RequestTimeout.

Fix

Pass an explicit per-call timeout to callTool, configurable via a new env var GLEAN_REMOTE_TOOL_TIMEOUT_MS:

• Positive integer; default 300000 (5 min), mirroring HITL_TIMEOUT_MS.
• Falls back to the default on unset / invalid / non-positive values.
• A single predictable knob — the env var is the timeout (no resetTimeoutOnProgress), so its name means exactly what it says. Easy to extend later for progress-streaming tools if needed.

Changes

• src/remote-client.ts — add exported remoteToolTimeoutMs(); thread the timeout into callTool's request options (resultSchema passed as undefined to keep the SDK default).
• README.md — document the new env var in the config table.
• tests/remote-client.test.ts — cover remoteToolTimeoutMs() (default, override, non-numeric, zero/negative) and callRemoteTool (timeout passthrough + contentless-result normalization).
• tests/find-skills.test.ts — update assertions for the new three-arg callTool signature.

Plugin manifests auto-bumped to 0.2.38 by the pre-commit hook.

Verification

• tsc --noEmit clean
• Full suite: 185/185 passing
• Confirmed GLEAN_REMOTE_TOOL_TIMEOUT_MS is present in the rebuilt plugins/glean/dist/index.js

Usage

Set in the host's MCP env block or shell, e.g. GLEAN_REMOTE_TOOL_TIMEOUT_MS=600000 for a 10-minute ceiling.