feat(router): reject oversized prompts before backend

[phaedrus-on-git]

Summary

• add alias-level max_prompt_chars config and advertise it in model cards
• reject oversized prompts before queue acquisition, ensure hooks, or backend proxying
• return OpenAI-shaped context_length_exceeded errors with compact-context headers and prompt-limit metrics

Tests

• go test ./internal/config ./internal/server
• make check