fix(mcp): respect HERMES_MCP_DISCOVERY_TIMEOUT env var for slow-connecting servers

[arimu1]

Summary

• hermes -z (headless oneshot) snapshots its MCP tool list once after only 0.75 s, so any MCP server with a real process startup (Docker, stdio) is silently absent for the entire session — no error, no warning.
• This PR adds a HERMES_MCP_DISCOVERY_TIMEOUT environment variable that overrides the discovery wait, giving users with slow servers a zero-config-schema escape hatch.

Root Cause

wait_for_mcp_discovery(timeout: float = 0.75) in hermes_cli/mcp_startup.py is called with no argument at both cli.py:793 and cli.py:4937. The 0.75 s default predates process-backed MCP servers. In headless mode the tool list is snapshotted once at startup and never refreshed.

Fix

wait_for_mcp_discovery now reads HERMES_MCP_DISCOVERY_TIMEOUT (float seconds) from the environment and uses it as the effective timeout when present. The existing 0.75 s default is preserved as the fallback — no call sites change, no config schema changes.

HERMES_MCP_DISCOVERY_TIMEOUT=10 hermes -z "call my-docker-tool and summarise the result"

Test Plan

• Configure a stdio/container MCP server whose connect time exceeds 0.75 s.
• Confirm hermes -z "list available tools" without the env var omits the server's tools (reproduces [Bug]: -z/headless oneshots silently drop slow-connecting MCP servers (0.75 s discovery wait too short for container/stdio-backed servers) #37013).
• Confirm HERMES_MCP_DISCOVERY_TIMEOUT=10 hermes -z "list available tools" includes the server's tools.
• Confirm interactive hermes (no -z) is unaffected.
• Confirm malformed value (HERMES_MCP_DISCOVERY_TIMEOUT=abc) silently falls back to 0.75 s.

Closes #37013

[teknium1]

Thanks for isolating the slow-MCP startup case and adding focused resolver tests.

This automated hermes-sweeper review is closing this as not_planned under the env-var-for-config policy.

• The proposed HERMES_MCP_DISCOVERY_TIMEOUT is a new user-facing, non-secret behavioral timeout override. AGENTS.md:102-105 requires behavioral settings such as timeouts to live in config.yaml, not a HERMES_* environment variable.
• Current main already provides the supported mechanism: mcp_discovery_timeout in hermes_cli/config.py:1359-1372, resolved by hermes_cli/mcp_startup.py:53-71 for the CLI and TUI paths.
• Current main also performs cache-safe between-turn MCP refresh in agent/turn_context.py:176-202 for servers that miss the bounded startup wait.

For a longer initial wait, configure mcp_discovery_timeout in config.yaml; no new environment-variable surface is needed.

---

Closed as not-planned per standing maintainer policy (env-var-for-config). This is a design-direction decision, not a code-quality judgment — see the Contribution Rubric in AGENTS.md for what the project is looking for. If you believe this policy was misapplied to your change, comment here and a maintainer will take a look.