Fix csvformat header options on empty input

[umd0730]

csvformat --skip-header and csvformat --no-header-row raise StopIteration on a zero-byte input. The same failure happens if --skip-lines consumes the whole input. This breaks pipelines whose upstream filter produces no rows.

Handle exhaustion when peeking for generated headers and when skipping an optional header. Empty input now produces no output; existing nonempty and blank-record behavior is unchanged.

Validation on Windows / Python 3.12.14 with agate 1.14.2:

• Before the fix, both added regression tests failed with StopIteration.
• After the fix, PYTHONUTF8=1 python -m pytest tests -q --tb=short --disable-warnings: 357 passed.
• Tests exercise all three combinations of --skip-header / --no-header-row, on zero-byte stdin and after skipping every input line.
• check-manifest, flake8 . and isort . --check-only all passed locally.
• Without UTF-8 mode, 14 unrelated tests fail reading UTF-8 expected files with the Windows CP932 default. The passing run uses Python's UTF-8 mode; no unrelated encoding changes are included.

Includes a changelog entry. Tested against current master ba8033dcbb2c72089e88f38a1a50ddfc8e5ac3df.