feat(data-quality): fill Flexoptix vendor_compat (coding matrix) from family templates

Investigated the coding-matrix gap per Rene: vendor_compat is NOT derived
per-SKU from fx_compatibilities (found rows with vendor_compat fully populated
but fx_compatibilities empty, and vice versa -- zero correlation). It is a
shared OEM-coding-pattern reference template keyed by (speed_gbps,
form_factor): only 11 distinct templates exist across the whole Flexoptix
catalog, essentially 1:1 with family (one ambiguous exception, 100G/SFP,
deliberately excluded). Backfilled 2,587 rows from sibling templates (fill-only,
skips any family with 0 or >1 distinct templates). Flexoptix >=100G coding gap
dropped from 192 to 25 missing. New trigger (trg_fill_vendor_compat_from_family)
keeps this self-healing: every future Flexoptix product inherits its family
template automatically once speed_gbps+form_factor are known (after detail
sync), instead of needing another one-off backfill.

Deliberately NOT extended to competitor products in this pass -- needs a
separate, more careful evaluation of what this data actually represents for
non-Flexoptix parts before propagating via equivalence.

sql/143-vendor-compat-template-fill.sql

@@ -0,0 +1,87 @@
+-- sql/143-vendor-compat-template-fill.sql
+-- Fills transceivers.vendor_compat (Flexoptix OEM-coding-pattern reference data,
+-- feeds v_coding_coverage / check_switch_quality's "coding matrix" metric) for
+-- Flexoptix products that are missing it.
+--
+-- Investigated 2026-07-17 (Rene: "die sind in der API alle zum auslesen drinnen"):
+-- vendor_compat is NOT derived per-SKU from fx_compatibilities (found rows with
+-- vendor_compat fully populated but fx_compatibilities completely empty, and vice
+-- versa -- no correlation). It is a SHARED REFERENCE TEMPLATE keyed by
+-- (speed_gbps, form_factor): only 11 distinct templates exist across the entire
+-- Flexoptix catalog, and every (speed, form_factor) family maps to exactly one
+-- template except 100G/SFP (2 templates -- ambiguous, deliberately excluded
+-- below). This was clearly seeded once (likely seed-from-npm.ts) for a subset of
+-- products and never propagated to new SKUs added since -- 167 of 192 missing
+-- rows (87%) have an unambiguous sibling template available to copy.
+--
+-- Fill-only, idempotent, no-false-positive design matching this project's other
+-- migrations: only touches rows with an EMPTY vendor_compat, only copies a
+-- template when exactly one distinct template exists for that (speed,
+-- form_factor) family among Flexoptix's own catalog, never invents new data.
+-- Competitor vendors are explicitly OUT OF SCOPE here -- see accompanying memory
+-- for why blind propagation to competitor products needs a separate, more
+-- careful pass.
+
+with templates as (
+  select speed_gbps, form_factor, (array_agg(vendor_compat))[1] as template
+  from transceivers t join vendors v on v.id = t.vendor_id
+  where v.name = 'Flexoptix' and jsonb_array_length(coalesce(t.vendor_compat,'[]'::jsonb)) > 0
+  group by speed_gbps, form_factor
+  having count(distinct vendor_compat::text) = 1
+)
+update transceivers t
+   set vendor_compat = tpl.template,
+       updated_at = now()
+  from vendors v, templates tpl
+ where v.id = t.vendor_id
+   and v.name = 'Flexoptix'
+   and tpl.speed_gbps = t.speed_gbps
+   and tpl.form_factor = t.form_factor
+   and jsonb_array_length(coalesce(t.vendor_compat,'[]'::jsonb)) = 0;
+
+-- Durable fix: new/updated Flexoptix products inherit the family template
+-- automatically once speed_gbps + form_factor are known (i.e. after the detail
+-- enricher runs), instead of relying on another one-off backfill.
+create or replace function tip_fill_vendor_compat_from_family() returns trigger
+language plpgsql as $$
+declare
+  fx_vendor_id uuid;
+  tpl jsonb;
+  tpl_count int;
+begin
+  if jsonb_array_length(coalesce(NEW.vendor_compat,'[]'::jsonb)) > 0
+     or NEW.speed_gbps is null or NEW.form_factor is null then
+    return NEW;
+  end if;
+
+  select id into fx_vendor_id from vendors where upper(name) like '%FLEXOPTIX%' limit 1;
+  if NEW.vendor_id is distinct from fx_vendor_id then
+    return NEW;
+  end if;
+
+  select count(distinct vendor_compat::text), (array_agg(vendor_compat))[1]
+    into tpl_count, tpl
+  from transceivers
+  where vendor_id = fx_vendor_id
+    and speed_gbps = NEW.speed_gbps
+    and form_factor = NEW.form_factor
+    and jsonb_array_length(coalesce(vendor_compat,'[]'::jsonb)) > 0
+    and id is distinct from NEW.id;
+
+  if tpl_count = 1 then
+    NEW.vendor_compat := tpl;
+  end if;
+  return NEW;
+end $$;
+
+comment on function tip_fill_vendor_compat_from_family() is
+  'BEFORE INSERT/UPDATE on transceivers: for Flexoptix products with empty vendor_compat, copies the OEM-coding-pattern template from an existing sibling of the same (speed_gbps, form_factor) family, if exactly one distinct template exists for that family. Never invents a template, never overwrites existing data, skips ambiguous families (>1 template).';
+
+drop trigger if exists trg_fill_vendor_compat_from_family on transceivers;
+create trigger trg_fill_vendor_compat_from_family
+  before insert or update of speed_gbps, form_factor, vendor_compat on transceivers
+  for each row execute function tip_fill_vendor_compat_from_family();
+
+insert into _migrations (filename, applied_at)
+select '143-vendor-compat-template-fill.sql', now()
+where not exists (select 1 from _migrations where filename = '143-vendor-compat-template-fill.sql');