Fix issues #45, #46, #47, #48

[powderluv]

Summary

• spur attach connects to job but does not provide interactive I/O #45: sattach now provides interactive I/O via bidirectional streaming AttachJob RPC (spawns shell in job namespace with stdin/stdout forwarding). Old output-only mode available via --output-only flag.
• spur nodes fails with "transport error / connection reset" when using custom ports #46: CLI reads controller address from /etc/spur/spur.conf (or $SPUR_CONF) when SPUR_CONTROLLER_ADDR is not set. Single change in main.rs sets env var before subcommand dispatch — all 14 CLI commands automatically inherit.
• Reopen #43 : Jobs still remain in PENDING state with Reason=Priority despite idle nodes #47: Nodes that don't match any partition hostlist are auto-assigned to the default partition on registration. Fixes jobs staying PENDING with Reason=Priority when dynamically-registered nodes had empty partition lists.
• Reopen #41: Container image still not found on agent despite successful import and listing #48: Agent's resolve_image() now tries basename fallback when absolute path from login node doesn't exist locally. Also adds ~/.spur/images fallback. Handles non-shared filesystem deployments.

Test plan

• 730 tests pass, 0 failures
• cargo fmt --check clean
• New tests: t50_92 (attach proto), t50_93 (config parsing), t50_94 (partition auto-assign), t50_95 (image basename fallback)
• CI: fmt, clippy, build-and-test, cluster tests

Closes #45, #46, #47, #48

🤖 Generated with Claude Code