docs: add mtp to accepted training.strategy values

[imitater-dou]

Motivation

examples/configs/README.md documents the accepted training.strategy
values as eagle3, peagle, dflash, domino, or dspark. The builtin
algorithm catalog registers six algorithms, including mtp:

# specforge/algorithms/builtin.py
return AlgorithmRegistry((eagle3(), peagle(), dflash(), domino(), dspark(), mtp()))

mtp is not an experimental name either — the checked-in recipe
examples/configs/online/disaggregated/managed-local/qwen3.5-4b-mtp-disaggregated-npu.yaml
already sets strategy: "mtp", and docs/basic_usage/training.md documents
MTP in its strategy/attention-backend notes and its
scripts/merge_mtp_to_base.py deployment flow. Readers who pick their
strategy from the recipe-catalog table therefore see one fewer strategy than
the code accepts.

Verified against main @ 6c6b75e:

$ python -c "from specforge.algorithms.builtin import builtin_algorithm_registry as r; print(sorted(r().names))"
['dflash', 'domino', 'dspark', 'eagle3', 'mtp', 'peagle']

$ python -c "from specforge.cli import load_config; print(load_config('examples/configs/online/disaggregated/managed-local/qwen3.5-4b-mtp-disaggregated-npu.yaml', []).training.strategy)"
mtp

An unknown value is rejected with the same list, confirming these are the
accepted values:

KeyError: "unknown algorithm 'bogus'; registered algorithms: ['dflash', 'domino', 'dspark', 'eagle3', 'mtp', 'peagle']"

Modifications

• examples/configs/README.md: add mtp to the training.strategy row's
  accepted values.

Docs only; one line. No strategy-specific row is added to the
"Strategy-specific fields" table because mtp defines no extra training.*
fields.

Related Issues

None.

Accuracy Test

Not applicable — documentation only, no model-side code touched.

Benchmark & Profiling

Not applicable — documentation only, no runtime code touched.

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.