Add toollog option and record MCP tool usage

[nmaguiar]

Motivation

• Provide dedicated, structured observability for MCP tool calls (inputs, outputs, cache and error flags) comparable to the existing auditch interaction audit.
• Make tool-level tracing available in both streaming and non-streaming runs so troubleshooting tool flows is straightforward.
• Expose the option across CLI/web schemas and docs so users can enable tool logging the same way they enable auditch.

Description

• Added a new runtime flag toollog and a _toollogon internal flag that initializes a _mini_a_toollog_channel from a JSSLON channel definition when supplied.
• Implemented MiniA.prototype._logToolUsage(toolName, params, answer, meta) to persist tool events (timestamp, tool, params, answer, connection id, cache flag, error flag) to _mini_a_toollog_channel when enabled.
• Integrated tool logging into the MCP execution path (_executeToolWithCache) to record cache hits, circuit-open failures, call exceptions, and successful results so all call outcomes are captured.
• Exposed and validated the toollog option in mini-a.yaml and mini-a-web.yaml, added the CLI/UI parameter in mini-a-con.js, and documented it in USAGE.md.

Testing

• Ran static syntax check with node --check mini-a.js which succeeded.
• Attempted to run the automated utility suite with ojob tests/miniAUtils.yaml but the ojob command is not available in this environment, so those tests were not executed here.
• Basic smoke checks: the modified code was inspected and the new logging calls are placed at cache hit, error, circuit, and normal-success locations in _executeToolWithCache to ensure coverage across call paths.