fix(teams): clamp bid award scores to 0-10

[BlueX888]

Summary

parse_bid_award wrote TL verdict scores straight to bid.score without the 0–10 clamp that parse_bid_scores applies. An out-of-range verdict score (e.g. 12.0 or -3.0) therefore flowed into the persisted bid record and into candidate_hp, which computes HP loss as 10 - score (see service.py:746) — an over-10 score could push an already-eliminated candidate's HP back above zero.

Why

The same bid.score field is written from two parsing paths:

• parse_bid_scores clamps: score: min(10.0, max(0.0, score)) (line 239)
• parse_bid_award does not clamp (line 269)

Only the clamp-on-scores path is exercised in normal flow; the award path still fills in any bid that missed a round score, so the inconsistency is reachable. The fix aligns both paths so the score's 0–10 domain holds everywhere.

Changes

• backend/app/teams/service.py: clamp parse_bid_award scores to [0, 10], matching parse_bid_scores.
• backend/tests/test_teams_bidding.py: new regression test asserting 12.0 clamps to 10.0 and -3.0 clamps to 0.0.

Testing

• pytest backend/tests/test_teams_bidding.py — 23 passed (includes the new test).
• pytest backend/tests/test_teams_api.py tests/test_teams_conversations.py tests/test_teams_ops.py tests/test_teams_bidding.py — 90 passed.
• ruff check backend/app/teams/service.py — clean.
• Sabotage check: removing the clamp makes the new test fail; restoring it makes it pass.

Related: #229 (its Q3 asks about the recommended way for a delivery manager to see member-task status; this fixes one defect in that area — an in-range score contract for team bidding).