* docs(README): point at the 'Help your AI find the tools' section

Brief callout + per-client rule-file table with a link to the full
canonical text in the install doc. Same problem we hit ourselves in
Cursor: agent has the MCP tools but doesn't pick them for natural-
language phrasings ('am I logged in?' triggers shell probes instead of
log10x_login_status). Project rules in the system prompt are the most
reliable lever; covered for Cursor, Claude Code, Windsurf, and Copilot.
Claude Desktop has no project-rule mechanism — flagged that explicitly.

* Add account/setup/discovery routing to MCP instructions

The server-level instructions block (injected into the LLM system
prompt by conformant clients) had thorough coverage for cost,
investigate, retriever, and root-cause workflows but no section for
auth/setup/status questions. Result: agents missed log10x_login_status
for natural-language phrasings like 'whats my login status' and
shelled out to which/env probes instead.

Adds explicit phrasing→tool mappings for login_status, signin,
signout, rotate_api_key, doctor, env CRUD, update_settings, plus a
hard 'no shell binary, MCP is the surface' anti-shell-probe rule
covering 'is the log10x CLI installed' / 'log10x version' framings.

Bumps to 1.6.7.

---------

Co-authored-by: Dor Levi <11015255+dor-levi@users.noreply.github.com>