Use native implementation of difflib

[JonoYang]

Using pyinstrument and scanning the the test file meta-quest-oss-notice.md for licenses, we have found that a lot of time is spent in the sequence matching portion:

11.902 get_licenses  scancode/api.py:150
├─ 11.858 detect_licenses  licensedcode/detection.py:2180
│  ├─ 9.619 LicenseIndex.match  licensedcode/index.py:892
│  │  ├─ 9.469 LicenseIndex.match_query  licensedcode/index.py:960
│  │  │  ├─ 8.915 LicenseIndex.get_approximate_matches  licensedcode/index.py:718
│  │  │  │  ├─ 6.753 LicenseIndex.get_query_run_approximate_matches  licensedcode/index.py:808
│  │  │  │  │  ├─ 6.360 match_sequence  licensedcode/match_seq.py:48
│  │  │  │  │  │  ├─ 5.469 match_blocks  licensedcode/seq.py:107
│  │  │  │  │  │  │  ├─ 5.446 find_longest_match  licensedcode/seq.py:19
│  │  │  │  │  │  │  │  ├─ 5.289 [self]  licensedcode/seq.py
│  │  │  │  │  │  │  │  ├─ 0.112 dict.get  <built-in>
│  │  │  │  │  │  │  │  └─ 0.045 extend_match  licensedcode/seq.py:84
│  │  │  │  │  │  │  │     ├─ 0.037 [self]  licensedcode/seq.py
│  │  │  │  │  │  │  │     └─ 0.008 <lambda>  <string>:1
│  │  │  │  │  │  │  │        ├─ 0.007 [self]  <string>
│  │  │  │  │  │  │  │        └─ 0.001 tuple.__new__  <built-in>

Currently, we use a pure python implementation of difflib to perform license detection (https://github.com/aboutcode-org/scancode-toolkit/blob/develop/src/licensedcode/seq.py) A way to improve the performance of this part of license detection would be to use a native implementation of difflib like https://pypi.org/project/cdifflib/ or https://github.com/rapidfuzz/CyDifflib

[armijnhemel]

The comments say that it is heavily modified. This makes me wonder:

1. are the modifications actually needed, or still needed?
2. would switching to the default difflib make sense?
3. if the modifications are needed, would these also need to be made to the difflib alternatives?