Add served model name option for server discovery

[RiccardoFiorentini]

Problem

Fixes #452. ds4-server currently advertises fixed DeepSeek-compatible model
IDs from the discovery endpoints, which can collide with real cloud-provider IDs
in multi-provider clients.

Change

• Add --served-model-name NAME.
• Validate NAME as non-empty and without /.
• Use the configured ID for GET /v1/models.
• Use the configured ID for GET /v1/models/<id>.
• Preserve current discovery behavior when the option is absent.

Non-goals

• No incoming request model validation or rewriting.
• No thinking-alias changes.
• No model loading, backend, inference, or performance changes.

Validation

Run on arm64 macOS 26.3 build 25D125, 32 GiB RAM, default Metal build.
Model/quant: not used.

• make clean && make && make ds4_test && ./ds4_test --server — passed.
• ./ds4-server --help | rg -F -- '--served-model-name NAME' — passed.
• make ds4_test && ./ds4_test --server && git diff --check — passed during cleanup review.

Also tried make test: Q4K unit tests, extractor self-tests, and agent tests
passed; the long-context stage stopped because ds4flash.gguf is not present.

No CUDA, ROCm, distributed, high-RAM, model-dependent inference, or performance
validation was run because this change is limited to server discovery metadata.