Add a Duration type and ARG_SECONDS to TtsSpan

This will enable more understandable speech output of durations
for screen readers, since by default numbers like "4:33" will
typically be announced as times, but depending on the context the
ideal speech may instead be "4 minutes 33 seconds".

This will also move developers away from content descriptions,
which reduce app resources used and avoids conflicts between
accessibility events (Example: simultaneous content description
and text changes in stopwatches/timers).

Flag: com.android.text.flags.tts_span_duration
Bug: 337103893
Test: atest TtsSpanTest, manual

Change-Id: I7e08eeaa67b0d1dbcf31dacf1465cb427dfa80c6