draft: memory: enhanced token-based memory management

[tmc]

This PR proposes some ideas for memory management, related to Discussion #124 and Discussion #317 about context length handling.

Configuration Example

memoryBuffer := memory.NewEnhancedTokenBuffer(
    memory.WithTokenLimit(2000),                    // Maximum tokens to keep
    memory.WithEncodingModel("gpt-3.5-turbo"),     // Model for token counting
    memory.WithTrimStrategy(memory.TrimOldest),     // How to trim messages
    memory.WithPreservePairs(true),                 // Keep human-AI pairs together
    memory.WithMinMessages(2),                      // Minimum messages to preserve
)

comments welcome.