fix(bedrock): cache system prompt in auto mode

[arielnabavian]

Summary

CacheConfig(strategy="auto") previously only wrote a cache breakpoint on the last user message, leaving the (usually largest and most static) system prefix uncached. For workloads that share a system prompt across calls with varying user messages, every call wrote to cache and none read — a net cost regression versus no caching.

Appends a cachePoint to the system prompt when caching resolves to "anthropic" for the model and the caller hasn't already placed one at the end. Both SDKs mirror the change. TS adds a systemTTL knob to BedrockCacheConfig to match toolsTTL/messagesTTL.

Fixes #3144. First of a 9-PR stack that closes #2970.

Stack

This is 1/9 in a stack that adds caching to every Strands model provider that can support it, culminating in default-on caching for BedrockModel (PR 10).

1. fix(bedrock): cache system prompt in auto mode ← this PR
2. feat(litellm): route cache_config per underlying
3. feat(openai): cache_config + prompt_cache_key
4. feat(llamaapi): cache_config + prompt_cache_key/retention
5. feat(mistral): cache_config + prompt_cache_key
6. feat(openai-responses): prompt_cache_options for GPT-5.6
7. feat(vercel): accept cacheConfig, warn on per-block-marker providers
8. feat(gemini): cache_config no-op acceptance
9. feat(bedrock): default cache_config to auto

Coordinated with — not duplicating — open PR #3571 (AnthropicModel cache_config + cache_tools).

Test plan

• All 189 Python bedrock tests pass
• All 178 TS bedrock tests pass
• mypy, ruff, prettier, eslint clean
• Real-API smoke test on Claude Sonnet 4.5 + Opus 4.6 + Haiku 4.5 (owner to run before merge)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!