v0.1.0

Added

• toolism measure: token cost of every tool in OpenAI (Chat and Responses), Anthropic or MCP
  tool JSON, or of a live stdio MCP server (--mcp), as a table, Markdown or JSON.
• toolism slim: validation-equivalent rewriting with nine safe rules (SL001–SL009) and two
  lossy annotation rules (SL101, SL102, --level annotations). The output keeps the input's
  format and envelope; per-rule and per-tool savings go to stderr.
• --verify: bidirectional equivalence checking of every changed schema. Annotation-only changes
  are proved; other changes are searched for counterexamples with hypothesis-jsonschema, and any
  tool with one is rolled back.
• toolism check: CI gate against a committed baseline with growth budgets
  (--max-increase 10% and/or tokens), --max-total and --max-tool. Writes a Markdown summary
  to $GITHUB_STEP_SUMMARY.
• toolism rules: the rule catalogue as a table, JSON, or the generated docs/rules.md.
• Tokenizers: offline heuristic, tiktoken:<encoding> (default o200k_base) and exact
  anthropic:<model> via the count_tokens API.
• Python API: load, parse, measure, slim, dump, slim_tools (cached, for agent
  code), list_rules, load_mcp_stdio.
• GitHub Action (uses: rakeshnitb23/toolism@v0.1.0) with status, total, delta and
  summary-file outputs.
• Benchmark on 14 real tool sets (159 tools): 37.4% fewer tokens in total, median 11.3% per tool
  set, with zero equivalence counterexamples.