Add deep search agents with context management

[muthukumaranR]

Summary

Add content condensation and context management capabilities to deep search agents for handling large search results.

Changes

• Content Condensation Component: LLM-based component that extracts only information relevant to research questions
• Research Synthesis Updates: Integrated content condensation with token budget allocation
• Deep Search Agent Updates: Added condensation before synthesis to prevent context overflow
• Token Management: Accurate token counting with tiktoken and intelligent allocation per result
• Debug Logging: Track token usage and compression ratios

Usage

# Content condensation component
from akd.agents.search.components.content_condensation import ContentCondensationComponent

condensation = ContentCondensationComponent()
result = await condensation.arun(ContentCondensationInputSchema(
    research_question="What are the effects of climate change?",
    search_results=search_results,
    max_tokens=4000
))

# Deep search with automatic context management
from akd.agents.search.deep_search import DeepSearchAgent

agent = DeepSearchAgent(debug=True)
result = await agent.arun(DeepSearchInputSchema(
    research_question="Climate change impacts on agriculture",
    max_search_results=20
))

Testing

• Integration tests for content condensation workflow
• Token counting and compression ratio validation
• Deep search agent context management tests