format_list selects the first existing style even when that style only defines some pattern parts. This raises KeyError for valid inputs such as German unit and French standard-narrow lists with three items. When 2 is missing but start and end exist, two-item lists can repeat the last item: Wolof unit-narrow formats ["A", "B"] as "A B B".

Resolve each missing part through the fallback styles, preserving the requested style's parts and collecting the result in a fresh dictionary so cached locale data is unchanged. Correct the unit-narrow chain to unit-short then standard, matching the root aliases in CLDR 45 and 48.2. The previous extra unit step could select wide-unit parts; removing it also changes Urdu two-item narrow lists from "A، B" to "A اور B" (including ur_PK and ur_IN).

Related to #781 and the whole-style fallback added in #1099 for #1098. #781's original German standard-narrow example already works on current master; this addresses the remaining partial-style failures and missing-two-item-pattern behavior.

Validation with generated CLDR 48.2 data:

• Full suite on Python 3.12 and 3.14, each with TZ=UTC: 7,846 passed, 7 skipped, 2 xfailed (baseline: 7,831 passed).
• List/unit tests: 49 passed. New regressions were demonstrated failing before the fix; the Santali case also failed before correcting the fallback entry.
• Repository pre-commit hooks passed for all files.
• External sweep of all generated locales, nine styles, and zero through five items: 60,534 cases, zero exceptions or differences from the root-alias oracle, unchanged cached patterns. All 3,488 baseline exceptions were repaired. Exactly the 100 prior successful outputs that differed from the oracle changed: 97 duplicated-item cases and the three Urdu two-item cases. All other successful outputs stayed identical.

The local timezone tests fail on the untouched base without TZ=UTC; the same timezone setting was used for the before/after full-suite comparison.