Reads up to max_read TextGrid files and returns the union of their
interval-tier names. Point tiers are ignored: landmarks come from interval
boundaries. Tiers are usually uniform across a corpus, so a small sample is
enough to populate a tier selector without reading thousands of files.
