Reproduce license cache inheritance failure #4709

[hakandilek]

Summary

Add a lightweight reproducer for the multiprocessing license-cache
inheritance failure reported in #4709, and a possible fix following a short
discussion.

Root cause: ScanCode warms the license cache in the parent process before
creating its worker pool. Linux Python 3.13 uses fork, so workers inherit
the warmed cache through copy-on-write. Python 3.14 uses forkserver by
default, so workers start without that state and load private copies of the
license index.

Reproducer

The test substitutes a small sentinel for the real license cache and creates
a worker through the production scancode.pool.get_pool() function.

This exercises the same inheritance boundary without:

• Loading the approximately 1 GB license index
• Measuring unstable RSS values
• Requiring a large input scan
• Consuming significant memory or runtime

Expected behavior on the current code:

Platform	Python	Start method	Result
Linux	3.13	fork	Pass
Linux	3.14	forkserver	Fail
macOS/Windows	Any	spawn	Skipped

The test is collected by the
core_tests / misc_and_scancode
job defined in azure-pipelines.yml, running on Ubuntu 24.04 with Python
3.14.

Scope

This Draft PR initially contains only the reproducer and no production fix.

I would like maintainer feedback on the preferred fix direction before
adding one.

Tests do not currently pass on Linux Python 3.14. This failure is
intentional while the PR is a Draft and demonstrates the regression.

Reproduces #4709.

Verification

venv/bin/pytest -vvs tests/scancode/test_pool.py