ICU-23335 Fix undefined behavior and null-deref in date_format_fuzzer

[OwenSanzas]

Summary

Fix two issues in icu4c/source/test/fuzzer/date_format_fuzzer.cpp:

1. Undefined behavior: invalid enum values

dateStyle and timeStyle are read from fuzz data via memcpy directly into EStyle enum variables without validation. The EStyle enum has only 13 valid values, so arbitrary 32-bit values are undefined behavior per the C++ standard.

Confirmed with UndefinedBehaviorSanitizer:

runtime error: load of value -559038737, which is not a valid value for type 'icu::DateFormat::EStyle'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior

The harness already has the correct pattern for dateStyle2/timeStyle2 — they use styles[rnd2 % numStyles]. The first pair was an oversight.

Fix: Use styles[rnd2 % numStyles] for all style variables, consistent with the existing pattern.

2. Missing null checks before df->format()

df->format(date, appendTo) is called without checking if df is nullptr after createDateTimeInstance. While createDateTimeInstance rarely returns nullptr in practice, the missing null checks are inconsistent with the later createInstanceForSkeleton calls which do check for success.

Fix: Add null checks (if (df)) before all df->format() calls.

Before/After Comparison (60-second fuzzing, AddressSanitizer)

Metric	Original	Fixed	Change
Edge cov	7106	8029	+13.0%
Features	21463	25788	+20.2%
Corpus	813	1052	+29.4%
Exec/s	375	380	+1.3%

The fixed version achieves 13% more edge coverage because valid EStyle enum values trigger more DateFormat internal code paths (e.g., RelativeDateFormat, various style combinations) instead of always falling through to the default fallback path.

Changes

• Read dateStyle/timeStyle as uint8_t + modulo index into styles[] (same as dateStyle2/timeStyle2)
• Adjust minimum input size accordingly (4 × uint8_t instead of 2 × uint8_t + 2 × sizeof(EStyle))
• Add null checks before all df->format() calls
• Reset status before udat_open call