bin: add lex-code-bootstrap wrapper; document --max-steps in the README

[alpibrupa]

Summary

• bin/lex-code already solves this for the TUI: the VM's default step budget (10,000,000, a DoS guard for untrusted sandboxed snippets) kills a long agentic session with step limit exceeded once enough turns or verbose-enough model output accumulate. bootstrap/run.lex — build/test/verify pipelines against a real task — is exactly the same "trusted, long-running orchestration" case, but had no wrapper of its own.
• Reproduced live, repeatedly (three separate lex-economy module builds this session): a real multi-module build running 10-20+ minutes across dozens of agent turns hit the panic with --max-steps omitted, discarding that whole run's output — bootstrap/run.lex only prints per-step output after the graph run returns, so a mid-run panic loses everything, not just the final result.
• Added bin/lex-code-bootstrap, mirroring bin/lex-code's own defaults and env-var overrides (LEX_CODE_EFFECTS, LEX_CODE_MAX_STEPS). Simpler than bin/lex-code: bootstrap/run.lex's main takes no positional args (everything is env-driven by the file's own design), so there's no main -- separator or "$@" forwarding to get right.
• Fixed the three bootstrap/run.lex invocation examples in the README's "Bootstrap Script" section, which was missing --max-steps even though the Quickstart and semantic-search sections both already document it for their own entry points — the exact gap that led to this session hitting the crash three times before finding the root cause.

Test plan

• Ran the wrapper against the repo's own zip demo task (LEX_PIPELINE=build) in a throwaway worktree — produced a correct, lex check-clean src/list.lex
• lex doc-sync --check passes
• git status scoped to exactly the two intended files

🤖 Generated with Claude Code