Add --min-word-length option to ignore short words

[ekanshul]

Closes #2211

Adds a --min-word-length LENGTH option so that words shorter than LENGTH are not checked. This helps with codebases full of 2-3 letter variable names that would otherwise each need an entry in --ignore-words-list.

Implementation: entries shorter than LENGTH are dropped from the misspelling dictionary right after it is built, so short words are ignored consistently everywhere (file contents, stdin and --check-filenames) with no per-word cost during scanning. The default (0) keeps current behavior.

Example:

$ codespell file.txt
file.txt:1: nwe ==> new
file.txt:1: abandonned ==> abandoned
$ codespell --min-word-length 4 file.txt
file.txt:1: abandonned ==> abandoned

Tests included (test_min_word_length) covering the length boundaries, case-insensitivity and --check-filenames.

[chuenchen309]

Applied this branch and exercised it rather than just reading it — it works, and filtering at the dictionary level is the right layer. Sharing the checks in case they're useful to a maintainer:

$ cat short.txt
The aas value and the abandonned config.

$ codespell short.txt
short.txt:1: aas ==> ass, as
short.txt:1: abandonned ==> abandoned

$ codespell --min-word-length 4 short.txt
short.txt:1: abandonned ==> abandoned          # short key dropped, long one still caught

The one thing I went looking for was whether len(key) could mean something other than "word length" — a multi-word dictionary key would make --min-word-length 4 compare against the whole phrase rather than a word. It can't: no builtin dictionary has a key containing a space (I checked all of codespell_lib/data/dictionary*.txt — zero). So len(key) is unambiguously the word length here. Worth knowing it holds by data rather than by construction, in case custom -D dictionaries are ever expected to carry phrases.

Filtering the dictionary once after build_dict rather than at each match site also means --check-filenames and the -w path get the same treatment for free, which matches what the PR description claims.

For context on the size of what it buys: 162 of the default dictionary's keys are ≤3 characters, so --min-word-length 4 is a meaningful reduction for a codebase full of short identifiers — the use case in #2211.

(Not a maintainer, just a contributor who was recently in this file.)

[ekanshul]

Gentle ping on this one — it's been about four weeks, so I wanted to check it hadn't just slipped off the radar.

One thing that may be worth a maintainer's attention: the GitHub Actions workflows on this PR have never actually run. All five runs on the head commit (Pytest, Build, codespell, autofix.ci) sit at completed/action_required, which is the first-time-contributor approval gate. pre-commit.ci is green, but since the required workflows never started, the PR reports as blocked on missing checks rather than on anything in the diff — so there's currently no way for me to show it passing from my side. Approving the workflow run should clear that.

It isn't specific to this PR: there are currently 42 runs in action_required across the repo, so a few other contributors are likely stuck the same way.

For context, this closes #2211, which is labelled good first issue, and @peternewman's last comment there invited anyone to pick it up. Thanks also to @chuenchen309, who applied the branch and verified the behaviour independently above.

The same gate applies to #3972 and #3973 if they're being triaged together.

Happy to rebase, split this up, or take a different approach if you'd prefer it done another way — just let me know.