RFR: 8313693: Introduce an internal utility for the Damerau–Levenshtein distance calculation [v3]

[Pavel Rappo]

> Please review this PR to introduce the Damerau–Levenshtein distance calculation for later use in REPL and CLI.
> 
> The Damerau–Levenshtein distance (DL) is a string similarity metric, variants of which have been successfully used in REPL and CLI:
> 
>     $ git hlpe
>     git: 'hlpe' is not a git command. See 'git --help'.
>     
>     The most similar commands are
> 	    grep
> 	    help
> 
> or
> 
>     $ hg clonmit
>     hg: unknown command 'clonmit'
>     (did you mean one of clone, commit, config?)
> 
> Originally, DL was requested by https://bugs.openjdk.org/browse/JDK-8288660 to provide helpful error messages for mistyped javadoc tags. However, given its wider applicability (e.g. mistyped command-line options), it was suggested that DL is put somewhere more internally accessible, for example, in `com.sun.tools.javac.util`, which is available in `jdk.compiler` and already exported to some other potential clients:
> 
>     exports com.sun.tools.javac.util to
>             jdk.jdeps,
>             jdk.javadoc,
>             jdk.jshell;
> 
> ---
> 
> 
> The Levenshtein edit distance is concerned with three types of edits: insertions, deletions, and substitutions. The Damerau-Levenshtein distance adds the fourth type: transpositions of two adjacent characters. This considerably complicates the algorithm, which is currently known in two main forms: restricted and unrestricted.
> 
> The difference between the two forms is that the simpler form, restricted, operates under the assumption that no substring is edited more than once. This results in different distances between some strings, for example, "CA" and "ABC". Restricted DL gives 3, whereas unrestricted DL gives 2.
> 
> Git seems to have [chosen](https://git.kernel.org/pub/scm/git/git.git/commit/?id=8af84dadb142f7321ff0ce8690385e99da8ede2f) restricted DL. This PR proposes **un**restricted DL. From my nonexpert perspective, while unrestricted DL does not necessarily guarantee a better end result, it feels a better place to start. Additionally, unrestricted DL is a proper metric. That property could probably be used to quickly estimate the upper bound of DL in some applications.
> 
> Like restricted DL, unrestricted DL runs in `O(s1.length() * s2.length())` in time. However, unrestricted DL requires more space: `O(s1.length() * s2.length())` plus the space for the alphabet, which for the purposes stated in this PR should be relatively small: `[a-zA-Z0-9]` plus maybe a dozen of additional symbols, such as `_` and `-`.
> 
> The algorithm that this PR proposes is my port of ...

Pavel Rappo has updated the pull request with a new target base due to a merge or a rebase. The incremental webrev excludes the unrelated changes brought in by the merge/rebase. The pull request contains four additional commits since the last revision:

 - Feedback: provide a usage example
 - Merge branch 'master' into 8313693
 - Feedback
 - Initial commit

-------------

Changes:
  - all: https://git.openjdk.org/jdk/pull/15157/files
  - new: https://git.openjdk.org/jdk/pull/15157/files/20db7d5d..5233e68d

Webrevs:
 - full: https://webrevs.openjdk.org/?repo=jdk&pr=15157&range=02
 - incr: https://webrevs.openjdk.org/?repo=jdk&pr=15157&range=01-02

  Stats: 9369 lines in 434 files changed: 3339 ins; 2874 del; 3156 mod
  Patch: https://git.openjdk.org/jdk/pull/15157.diff
  Fetch: git fetch https://git.openjdk.org/jdk.git pull/15157/head:pull/15157

PR: https://git.openjdk.org/jdk/pull/15157