Implement caching mechanism to speed-up formatting of Elm blocks in markdown files

[kachkaev]

Prettier parsers and printers work synchronously until prettier/prettier#4459 is discussed and resolved. This means that formatting a markdown file with a large number of Elm blocks takes a while, because for each block a new elm-format is spawned in a sequence.

This PR introduces a simple LRU caching mechanism that works on top of the filesystem. The cache speeds-up consequent formatting of the same text, even between independent prettier executions. Occasionally, garbage collection is invoked to protect cache directory from growing indefinitely.

Each new version of prettier-plugin-elm will use its own cached results, because the value of the version from package.json is mixed with the text to format when generating cache key.

By default, the cache is stored in $TMPDIR/prettier-plugin-elm, but this location can be changed with the environment variable. In practice, this is only needed during testing, so caching can be considered as zero-config.

Caching can be extracted into a separate NPM module as it's pretty generic and I'm happy to do so if someone is interested in reusing it (just ping me).