[Feature] Refactor configs dir

[jiapingW]

Motivation

The example training configs previously lived mostly in a flat directory and relied on inconsistent filename suffixes such as -online, -offline, and
-disaggregated. These names did not reliably describe the actual runtime topology or service ownership.

This PR reorganizes the recipe catalog around explicit runtime semantics, making configs easier to discover, understand, and maintain.

Modifications

• Reorganized all 68 example YAML configs into the following hierarchy:

  examples/configs/
  ├── offline/
  │   ├── colocated/
  │   └── disaggregated/
  └── online/
      ├── disaggregated/
      ├    └── external/
      ├    └── managed-local/
      └ colocated/
  

• Classified recipes according to their actual configuration fields:

  • offline/colocated: directly consumes precomputed features.
  • offline/disaggregated: uses separate producer and consumer roles.
  • online/disaggregated/external: connects to externally managed Mooncake and SGLang services.
  • online/disaggregated/managed-local: lets SpecForge manage local Mooncake and SGLang services.

• Added documentation for the currently unsupported online/colocated topology.

• Preserved the contents and training behavior of all migrated YAML recipes.

• Renamed AMD recipe files with an -amd suffix so platform variants fit the same topology hierarchy.

• Updated config discovery tests to recursively scan the new directory structure.

• Added validation ensuring every recipe is stored under the directory matching its configured mode, topology, and service ownership.

• Updated documentation, examples, scripts, tests, and operational replacement mappings to use the new paths.

Related Issues

N/A.

Accuracy Test

Not applicable. This PR only reorganizes configuration files and updates their references; it does not modify kernels, model architectures, training objectives,
or model-side numerical behavior.

Validation performed:

• 56 config tests passed.
• 52 launch-plan tests passed.
• 18 package-architecture tests passed in a clean working tree.
• All pre-commit checks passed.
• All 68 migrated YAML files were verified to have unchanged contents.
• All repository references to example config paths were checked successfully.

Benchmark & Profiling

Checklist

• [✅] Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://sgl-fru7574.slack.com/archives/C09784E3EN6 to discuss your PR.

[maocheng23]

LGTM, feel free to merge