Flag hotels absent from successful scrapes as 'not_listed' + widen booking_com_id

Location-search results aren't a fixed hotel set — a sold-out hotel drops
out and its last 'available' rate would remain the latest row for that
date, reading as a live price and skewing market averages. On each
successful per-date scrape, insert a NULL-rate 'not_listed' row for every
active hotel missing from the results (skipped if the parse found nothing,
which indicates scraper fault not absence). Failed/blocked scrapes write
nothing, so genuinely-stale data remains distinguishable by scraped_at.

Also: fix the DOW analysis to pick latest-then-filter so a not_listed
latest row drops the date instead of resurfacing an older rate, and widen
booking_com_id to VARCHAR(255) (some Booking slugs exceed 50 chars).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

backend/api/analysis.py

@@ -109,15 +109,17 @@ async def analyse_hotel(
                     ROUND(AVG(rate_gross)::numeric, 2)      AS avg_rate,
                     COUNT(DISTINCT rate_date)               AS date_count
                 FROM (
+                    -- Latest row per date first, THEN filter: a 'not_listed'
+                    -- latest row must drop the date, not resurface an older rate
                     SELECT DISTINCT ON (rate_date)
-                        rate_date, rate_gross
+                        rate_date, rate_gross, availability_status
                     FROM booking_com_rates
                     WHERE hotel_id  = :hid
                       AND rate_date BETWEEN :from_date AND :to_date
-                      AND availability_status = 'available'
-                      AND rate_gross IS NOT NULL
                     ORDER BY rate_date, scraped_at DESC
                 ) latest
+                WHERE availability_status = 'available'
+                  AND rate_gross IS NOT NULL
                 GROUP BY dow, dow_label
                 ORDER BY dow
             """),