feat: robust pattern-matching cost estimation and trace decorator resilience

[archittmittal]

This Pull Request resolves Issue #25, Issue #29, and Issue #11.

Key Changes:

1. Pattern-Matching Price Resolution (Issue Flawed LLM Cost Estimation for Versioned and Pinned Models #25):
   • Modified packages/sdk/agentscope/_pricing.py to resolve model cost estimations using prefix/pattern matching. Base pricing patterns are sorted by length in descending order to guarantee that more specific version-pinned names (e.g. gpt-4o-2024-05-13) match before generic ones.
   • Introduced programmatic configuration overrides (update_pricing_table) and environment variable configuration overrides via AGENTSCOPE_CUSTOM_PRICING parsed dynamically at runtime.
2. Claude 3 & Gemini 1.5 Pricing (Issue feat(sdk): Add support for Claude 3 (Haiku/Opus) and Gemini 1.5 Flash in pricing table #29):
   • Populated default price parameters in PRICING_TABLE for claude-3-haiku, claude-3-opus, and gemini-1.5-flash.
3. Trace Decorator Resilience (Issue feat(sdk): Implement @trace custom decorator instrumentation (Sub-task 2.4) #11):
   • Covered all client setup and websocket event emission logic in safety try/except blocks in packages/sdk/agentscope/decorators.py to ensure telemetry server connectivity errors never propagate and disrupt user-defined synchronous/asynchronous logic.
4. Verification:
   • Expanded packages/sdk/tests/test_sdk.py with test_decorators_resilience confirming telemetry emit failures do not halt execution, and added version-matching, overrides, and Claude/Gemini cost assertions to test_calculate_cost.

All tests pass successfully.