feat(config): route manager reward overrides

[TATP-233]

Summary

• resolve the Hydra root reward target from the registered env config owner
• route legacy configs to reward_config and explicitly marked Manager-Based configs to rewards
• reject unregistered/missing/ambiguous targets and duplicate root/env.rewards sources
• preserve Python-owned reward callables, sibling terms, insertion order, and legacy dict replacement semantics

Implements #1070.
Part of #1042.

Scope accounting

• source-derived code: 0 lines
• mechanical conversion: 0 lines
• UniLab-specific implementation/glue: 64 added, 5 deleted lines
• tests: 139 added lines
• files changed: 4

Validation

• focused routing tests: 6 passed
• registry/config/training/script regression: 204 passed
• make test-all:
  • ruff, mypy, pyright passed
  • pytest: 1927 passed, 25 skipped, 274 deselected, 1 xfailed
  • benchmark module mode: 32/33 passed, 1 platform-optional skip
  • benchmark script mode: 33/34 passed, 1 platform-optional skip