Self-hosted runner can snapshot and verify the operator's host config

What

The self-hosted runner (the claude runner binary used for hosted/CI-style session farms) can now capture a snapshot of the operator's host config directory at startup and seed it into every session's CLAUDE_CONFIG_DIR.

• The snapshot covers settings.json, settings.local.json, CLAUDE.md, hooks/, skills/, plugins/, agents/, and the .claude.json mcpServers entry.
• Mode is chosen with the new SELF_HOSTED_RUNNER_HOST_CONFIG_SNAPSHOT environment variable (also settable via --host-config-snapshot <disk|memory>), accepting disk or memory case-insensitively and defaulting to disk if unset or blank.
• In disk mode (the default), the config directory is copied once into a runner-owned directory and its file hashes are checked against an in-memory manifest at every session start, with no size limit; if tampering is detected, sessions fail and the runner must be restarted.
• In memory mode, the config is held on the heap, capped at 64 MiB; sessions silently start without host config once that cap is exceeded.
• The source directory defaults to ~/.claude, can be redirected or disabled (empty directory) via SELF_HOSTED_RUNNER_HOST_CONFIG_DIR, and a related CLAUDE_CODE_PLUGIN_SEED_DIR variable can pre-populate large plugin sets.

Why

This lets operators of self-hosted runner fleets give every session a consistent, verified copy of shared settings, hooks, skills, plugins and agents, while catching tampering with the on-disk snapshot before it can affect a session.