Pretty short change for reducing O(n^2) impact with a cache.

This change includes the following scalability improvements for documents containing extremely large paragraphs:

- Reduces the size of layout contexts to account for LF control chars.

- Due to typical access patterns while laying out paragraphs, VCL was making O(n^2) calls to vcl::ScriptRun::next(). VCL now uses an existing global LRU cache for script runs, avoiding much of this overhead.