Earlier entries in word boundary cache are cleared incorrectly

[milseman]

This fails, because we clear earlier word boundary cache entries when we calculate later ones.

    // This will first find word boundaries up til the middle before failing,
    // then it will find word boundaries til late in the string, then fail,
    // and finally should succeed on a word boundary cached from the first
    // attempt.
    let input = "first second third fourth"
    let regex = try Regex(#".*second\bX|.*third\bX|.*first\b"#)
    XCTAssertTrue(input.contains(regex))