docs(tasks): split 0038 lexical scoping into 0165-0167, file 0168/0169, add the approved plan

[rmorozov]

Requested by Ruslan · project thread

Before: 0038 (lexical scoping for macros) was one open task with no plan. Its six divergence lines named 0038, and the task's touches held render_context.h and the macro code in every task_batches.py wave.

After: the plan Ruslan approved is in docs/design/0038-lexical-scoping-plan.md, with its advisor review next to it. 0038 is now an umbrella task over three phase tasks, each owning the divergence lines it removes:

• 0165: macro and call bodies, loop() and super() run where they were defined. It owns 5 lines and can run in wave 2.
• 0166: defaults are evaluated per call. It owns macro_default_reassigned_global.
• 0167: imported modules own their context. The render owns each module and exports hold weak anchors, so no cycles form and no sweep is needed. It depends on 0155.

Two out-of-scope findings are filed and pinned with new corpus cases:

• 0168: Jinja2's idtracking binds a name at frame entry. The new case is statements.idtracking_inner_read_before_outer_set: Python renders '', we render 33.
• 0169: caller() without a call block should raise "No caller defined". The new cases are errors.caller_without_call_block and errors.call_body_reads_own_missing_caller.

How: this PR adds task files and corpus rows only, with no src/ changes. expected/*.json was regenerated with jinja2 3.1.6. I left out an unrelated operators.json message change that comes from the local Python version. ctest -R parity passes locally (2941 tests). docs/tasks/README.md is not edited; the steward runs task_index.py.

Roles: architect (plan), advisor (review). Verifier not run because there is no code change. Pushes red in CI: none so far.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Krh9zQy8ifSut6nW1jwRZf

---

Generated by Claude Code