Speed up import isal.igzip by not importing the typing module - #319

import isal.igzip and import isal.igzip_threaded also import the typing module, but the two modules use the names that they import from it only in annotations. typing in turn imports other modules, for example, functools on Python 3.14 and re, enum, and contextlib on Python 3.11, so every program that imports isal.igzip pays for them. python -m isal.igzip loads most of these modules anyway through argparse and shutil, so on the command line this PR saves only the import of typing itself.

With this PR, neither module imports typing:

• igzip.py annotates the mtime parameter of compress() as int | None instead of Optional[SupportsInt]. Type checkers now report an mtime that isn't an int, but at run time compress() still converts mtime with int().
• igzip_threaded.py uses list, tuple, and X | None in its annotations instead of List, Tuple, and Optional.

The test_import_does_not_load_cli_modules test is now test_import_does_not_load_slow_modules, and it also checks that typing isn't imported. That check fails on develop.

I measured the wall-clock time of new processes that run python -c "import isal.igzip", in paired rounds against develop. The Linux numbers are from Docker on WSL2, and the Windows numbers are from native builds. The table shows the medians in milliseconds:

On Linux with CPython 3.14, python -c pass takes 7.6 ms, so importing isal.igzip adds about 5.2 ms to the process on develop and 2.1 ms with this PR. On the five platforms in the table, python -m isal.igzip -c on a 64 KiB file finishes 1.3 to 2.4 ms sooner. A program that imports typing for other reasons saves nothing. xopen and dnaio both import typing, for example.

The test suite passes on this branch on Windows with CPython 3.14 and 3.11.

Checklist

• Pull request details were added to CHANGELOG.rst
• Documentation was updated (if needed)