The longest gap between listening days

Constraints

1. 1Collapse to distinct listening days first. A day with four plays is one listening day.
2. 2longest_gap_days is the largest number of days between two consecutive listening days for that listener, as a whole number.
3. 3A listener with a single listening day has no gap at all: report 0.
4. 4Listeners who have never played anything do not appear — this query reads stream_plays only.
5. 5Order by user_id.

Worked example

u3 listened on 9 and 10 January, then 14 February, then 8 March. The gaps are 1, 35 and 22 days, so listening_days is 4 and longest_gap_days is 35. Skip the distinct-day step and u3's six raw plays produce extra gaps of 0 between two plays in the same evening. Here that noise does not change the maximum — but it makes listening_days report 6 sittings as 6 days, which is simply wrong.

What this tests

Combining LAG with a deliberate change of grain: dedupe to the unit the question is really about, then compare each row with the previous one inside its partition.