fix(providers): clamp unsupported temperatures in Claude Code provider

[bjorn]

Summary

• The Claude Code CLI only supports temperatures 0.7 and 1.0, but internal subsystems (memory consolidation at 0.1, context summarizer at 0.2) request lower values
• Previously the provider rejected these with a hard error, triggering retries and WARN-level log noise via ReliableProvider
• Now clamps to the nearest supported value with a debug-level log, since the CLI ignores temperature anyway

Test plan

• Existing unit tests updated and passing (cargo test --lib providers::claude_code — 14/14 pass)
• Manual: run zeroclaw daemon with claude-code provider and verify no temperature warnings on incoming messages
• Manual: trigger memory consolidation and verify it completes without retry warnings

🤖 Generated with Claude Code

[bjorn]

I tried to fix these errors which I am seeing repeatedly when using the claude-code provider:

2026-03-19T09:53:09.382416Z  WARN zeroclaw::providers::reliable: Provider call failed, retrying provider="claude-code" model="default" attempt=1 backoff_ms=500 reason="retryable" error=temperature unsupported by Claude Code CLI: 0.1. Supported values: 0.7 or 1.0
2026-03-19T09:53:09.883611Z  WARN zeroclaw::providers::reliable: Provider call failed, retrying provider="claude-code" model="default" attempt=2 backoff_ms=1000 reason="retryable" error=temperature unsupported by Claude Code CLI: 0.1. Supported values: 0.7 or 1.0
2026-03-19T09:53:10.884933Z  WARN zeroclaw::providers::reliable: Exhausted retries, trying next provider/model provider="claude-code" model="default"

I've let Claude Code try to figure out a solution, but I realize this might not be the desired direction.

[theonlyhennygod]

LGTM. Clean approach to clamp unsupported temperatures to nearest supported value instead of hard-erroring, which unblocks callers like memory consolidation. Good test coverage.