Count Strands reasoningContent, image, document, video tokens (#111 follow-up)

Fix Strands SDK content block tokenization (#111)

Fix code review issues: batch NameError, null safety, token types

Fix image token estimate: use 1600 tokens (pixel-based), not 85

Fix image token counting: Anthropic images counted as text tokens

Add persistent memory system with zero-latency inline extraction

Fix all ruff lint and format errors for CI

Prepare for OSS release v0.2.0