Fix Japanese TN word spans with quoted spaces

[kiwigitops]

Fixes #427.

Summary:

• lets the Japanese word classifier consume spaces inside plain word spans by matching the existing Chinese TN word behavior
• keeps the quoted English phrase in the reported Japanese sentence verbalizable instead of leaving a failed lattice
• adds a Japanese word regression for the “Hello World” case

Verification:

• python -m compileall nemo_text_processing/text_normalization/ja/taggers/word.py tests/nemo_text_processing/ja/test_word.py
• git diff --check