Define sshx agent execution contract and close E2E quality gaps

[jamiesun]

Summary

• reposition sshx as an agent-native remote host execution tool: SSH is the trusted channel and X is execution
• define efficiency, security, product boundaries, future directions, and a mandatory capability acceptance matrix
• add compiled-binary E2E coverage across real SSH/SFTP protocol boundaries for commands, JSON/exit codes, permissions, partial completion, host trust, dry-run, host import, SFTP, server-to-server transfer, keyring-backed sudo, and audit recovery
• add Linux and macOS E2E CI, including an ephemeral macOS Keychain test for the production binary
• add canonical make test-short and make test-e2e workflows

Fixes

• correctly fall back to an explicitly supplied SSH password after public-key authentication is rejected
  • root cause: the previous code checked ssh.ServerAuthError, a server-side type that the real SSH client path does not return
  • the new E2E reproduces the real handshake and verifies both successful key authentication and password fallback
• surface audit persistence failures at error log level while preserving a successful remote command result

Impact

Agents now have documented, test-enforced guarantees around target resolution, structured outcomes, authentication, trust, permissions, side effects, and recovery. Existing implemented top-level capabilities meet the acceptance floor; unimplemented multi-host execution and explainable governance remain explicitly marked as future work.

Validation

• make check
• go test -short -race ./...
• make test-e2e
• production binary against a temporary macOS Keychain with default/search-list restoration verified
• make lint
• actionlint .github/workflows/ci.yml
• mdbook build
• git diff --check