This callback function was added for select pages in some namspaces,
this should be done with the `namespace_ids` arguemnt to query the
sqlite database.

Also fix the subprocess still running resouce warning.

The code doesn't reply on the Linux fork method to copy variables now.
multiprocessing code are moved to wiktextract, unpicklable objects
like sqlite connection and lupa runtime are handled at there.

This file is mostly test the `process_dump()`, move it to
`tests/test_dumppaser.py`. Extract each page is still tested in
wiktextract.

`get_page()` is kind slow(1s per call), cache the requests improve the
performance significantly. This reduces the process time of Chinese
Wiktionary from 40 minutes to 10 minutes.

…_template`

Using `pkg_resources` is not recommanded: https://setuptools.pypa.io/en/latest/pkg_resources.html

The English Wiktionary has 40327 template and 54524 module
pages(exclude redirects). I think the cache is mostly useful for
caching shared templat and module pages used in pages processes in a
worker process. So the cache size relates to page numbers, CPU core
numbers and shared templates. If 1000 is not enough, we can increase
it to 10000.