perf(tools): Simplified tool definition and schema generation, with the bonus that stateful tools are equally straightforward to introduce

feat: inference statistics (including TTFT, TPS, total inference time), computed by default

feat: automated logging (to file), and colorful {debug, warning, and errors printed to screen}, with removal of all but streaming-response-related print() statements

fix: proper parameterization of thinking (enabled/disabled, low/med/hi reasoning effort) and a simple CLI arg --minimize-thinking, and squashing of small bugs related to tool calling

feat: added support for running models served by llama.cpp, and in the process fixed some minor defects such as missing tool_call_ids in the chat message as well as expanded the example tools towards benchmarking model performance

feat: simple CLI (documented), leading to a code restructure (also documented) and fixes to a number of issues related to streaming and tool-calling, now with vLLM-served models fully tested and supported

feat: tools (both python functions and mcp-server-provided) are now in the flow, with weather and web crawling as examples in flat_mcp_client.tool_defs. Minimal 3-line example stands in for a more configurable main(), which can in turn be invoked by: geolocal_info

feat: basic structure, tested and working, with streaming chat loop standing in for a generalized flow, and tools and resources to come next

chore: README.md tagline tweak

Update README.md

Initial commit