APR-157: add hosted SFT CLI config

[tim0120]

Summary

Adds the hosted SFT distillation config surface to the Prime CLI.

• parses loss = "sft" from hosted training TOML
• parses [teacher] and [teacher.sampling]
• validates that SFT requires a teacher and RL forbids one
• keeps OPD gated until hosted teacher logprob scoring exists
• serializes loss and teacher into the /rft/runs payload
• updates the generated config template and focused config tests

This uses the newer GA-facing loss/teacher shape and is intended to supersede the older loss_type/teacher_rollout_model shape in PR #514.

Validation

• uv run python -m pytest packages/prime/tests/test_rl_config.py
• uv run ruff check packages/prime/src/prime_cli/commands/rl.py packages/prime/src/prime_cli/api/rl.py packages/prime/tests/test_rl_config.py
• git diff --check

Linear: APR-157