fuzz(asyoutypeformatter): use a valid region and phone-relevant input so the formatter is actually fuzzed

[sushant-me]

Summary

The fuzz_asyoutypeformatter target is currently ineffective: it builds and runs, but
barely exercises AsYouTypeFormatter. In the OSS-Fuzz coverage report the target file
sits at 0.00% (0/532) lines / 0.00% (0/29) functions:

https://storage.googleapis.com/oss-fuzz-coverage/libphonenumber/reports/20250202/linux/file_view_index.html

Root cause (two compounding defects)

1. Random region → empty metadata. The harness constructs the formatter with
   ConsumeBytesAsString(region_is_2_bytes ? 2 : 3), i.e. 2–3 arbitrary bytes that are
   essentially never a valid ISO-3166 alpha-2 code. AsYouTypeFormatter::GetMetadataForRegion
   then falls through to empty_metadata_ (zero NumberFormat entries), so
   possible_formats_ stays empty and no formatting template is ever built.

2. Random char32_t → able_to_format_ = false. Each character is
   ConsumeIntegral<char32_t>(), a uniformly random 32-bit code point that is essentially
   never a digit (or a leading +). InputDigitWithOptionToRememberPosition immediately
   sets able_to_format_ = false and the formatter only echoes the raw input, so the
   digit state machine (AttemptToExtractIdd, AttemptToExtractCountryCode,
   AttemptToChooseFormattingPattern, InputDigitHelper, …) is never reached.

Fix

• Select the region from a fixed set of real ISO-3166 codes so real metadata is loaded.
• Feed characters from a phone-relevant alphabet (0123456789+*#() -) instead of
  arbitrary char32_t, keeping InputDigit on the formatting path while the fuzzer
  still controls the exact sequence/length.

Measured effect

Built from this repo with clang 22 source-based coverage and ran 20 000 executions from
an empty corpus, measuring asyoutypeformatter.cc:

harness	line coverage	function coverage
before	23.50% (125/532)	31.03% (9/29)
after	93.05% (495/532)	93.10% (27/29)

No sanitizer report in either run.