[codex] Split dry-run resolved and effective config

[xu16601526267]

Summary

This PR makes deploy dry-run output distinguish catalog/user-resolved config from the final config that AIMA would actually deploy after CheckFit resource adjustments.

What changed

• keeps existing config as the effective deployment config for backward compatibility
• adds resolved_config and resolved_provenance for pre-CheckFit values
• adds effective_config, effective_provenance, and top-level fit_adjustments
• covers the behavior with a focused resolve/deploy test

Why

When CheckFit lowers gpu_memory_utilization or mem_fraction_static because of live memory pressure, UI and agents previously saw only the adjusted config. That made users think the catalog default itself was wrong.

Validation

• go test ./cmd/aima -run TestResolveDeploymentKeepsResolvedAndEffectiveConfigSeparate -count=1
• git diff --check

[xu16601526267]

Linux target validation passed on aitopatom-5ba7 (qujing@222.128.15.222:22222, Linux 6.11.0-1016-nvidia, aarch64) with temporary user-local go1.25.0 linux/arm64.

Validation run:

• git diff --check origin/develop...HEAD
• go test ./cmd/aima ./internal/knowledge ./internal/runtime -count=1
• go test ./... -count=1

Remote logs:

• related package logs: /home/qujing/aima-pr-ready-validation/logs-20260515T172839Z
• full package logs: /home/qujing/aima-pr-ready-validation/logs-full-20260515T172935Z