Complete temporary run retention coverage

[codeforester]

Summary

• add failed-run coverage for default temp cleanup and explicit retention
• document the per-run temporary-data contract in the observability guide
• keep run metadata, history, and the primary log available in both modes

Why

The retention implementation is already on main; this PR completes issue #1685's
remaining acceptance coverage and documentation. Temporary data is diagnostic-only
and should not accumulate after successful or failed commands unless the user
explicitly opts in with --keep-temp (or the documented Python retention setting).

Validation

• bats cli/bash/commands/basectl/tests/runtime-dispatch.bats — 29 passed
• BASE_BASH_LIBS_DIR=... env -u BASE_HOME ./bin/base-test — 1043 passed, 1 skipped
• PYTHONPATH=lib/python python3 -m unittest lib/python/base_cli/tests/test_base_cli.py lib/python/base_cli/tests/test_app_runtime_boundary.py — 49 passed, 14 skipped
• git diff --check
• ShellCheck reports only the existing SC2016 informational warnings in the Bats file

Project and AI context

• Base Project item moved to In Progress; merge/closure should move it to Done.
• .ai-context/ unchanged: this is completion coverage and documentation for an
  already-shipped runtime contract, with no new product or architecture decision.

Fixes #1685