Reduce dependencies for license scanning API

[shivd1410]

Fixes #5115

Summary

This PR reduces the dependencies loaded when using scancode.api.get_licenses().

Changes

• Removed the packagedcode.utils import from get_licenses().
• Import combine_expressions directly from license_expression instead.
• Moved typecode imports into the functions where file-type information is actually needed.
• Added a regression test to make sure importing get_licenses() does not load typecode, extractcode, or packagedcode.

Procedure

I started by tracing the imports pulled in when importing scancode.api.get_licenses().

• packagedcode.utils was pulling in a much larger package-scanning dependency tree, even though get_licenses() only needed combine_expressions.
• The top-level typecode import was also pulling in extractcode and other dependencies through its import chain.
• I checked where typecode was actually used and moved those imports to those specific functions instead of loading it when the module is imported.
• I then checked that the license scanning path still works without these dependencies being loaded upfront.

Testing

• Lightweight import regression test: passed
• get_licenses() functional check: passed (MIT detected correctly)
• pytest tests/scancode/test_api.py: 11 passed, 2 failed
• The 2 failures are in the existing licensedcode/cache.py boolean-expression handling and are unrelated to these changes.
• git diff --check: passed

CI

• Several CI jobs are currently failing in areas unrelated to this change, including
  ClueCode, misc/scancode, and cross-platform test suites.
• The same license_expression boolean-evaluation failure was also reproduced
  locally in the existing API tests and is unrelated to the changes in this PR.
• The new regression test for the lightweight get_licenses() import passes.