Make short-term memory context window configurable

[veithly]

Summary

• Make ShortTermMemoryConfig.max_tokens configurable from environment variables.
• Prefer SPOON_AI_SHORT_TERM_MAX_TOKENS, SPOON_BOT_CONTEXT_WINDOW, SPOON_CONTEXT_WINDOW, then CONTEXT_WINDOW.
• Raise the fallback default to 1_000_000 tokens so long tool-driven sessions are not compacted at the previous 8k default.

Validation

• Ran Python compile check for spoon_ai/chat.py.