Fix bytes serializer dropping non-UTF-8 values to null

[gaoflow]

The bug

Serializer.do_bytes serializes any non-UTF-8 bytes attribute to null, so values silently disappear from to_json / to_csv / to_dict output:

from pdfplumber.convert import Serializer, ENCODINGS_TO_TRY
print(ENCODINGS_TO_TRY)                    # ['utf-8', 'latin-1', 'utf-16', 'utf-16le']
print(Serializer().serialize(b"caf\xe9"))  # -> None   (expected 'café')

End-to-end, on the shipped tests/pdfs/issue-463-example.pdf (an annotation whose Contents is UTF-16BE):

import json, pdfplumber
pdf = pdfplumber.open("tests/pdfs/issue-463-example.pdf")
a = json.loads(pdf.to_json(object_types=["annot"]))["pages"][0]["annots"][0]
print(a["contents"], "|", a["data"]["Contents"])   # '日本語' | None   <- data loss

Root cause

def do_bytes(self, obj: bytes) -> Optional[str]:
    for e in ENCODINGS_TO_TRY:
        try:
            return obj.decode(e)
        except UnicodeDecodeError:  # pragma: no cover
            return None             # <-- aborts the loop on the first failure

The except returns None on the first encoding that fails, so the latin-1 / utf-16 / utf-16le fallbacks never run. git blame shows the type-annotation refactor in 9587cc7 changed the original pass (continue the loop) to return None. Because latin-1 decodes any byte string, the intended loop never returns None and never reaches the trailing re-raise — both of which became dead code (the # pragma: no cover even encodes the now-false assumption that the except never fires).

The fix

Restore the loop: continue to the next encoding instead of returning None (and drop the now-reachable # pragma: no cover). One line; the trailing re-raise is left untouched.

I deliberately do not change the encoding order — latin-1 precedes utf-16, so a UTF-16 value comes back as latin-1 text rather than its "ideal" decoding. That ordering is a separate, pre-existing design choice; restoring the loop exactly matches the original behaviour and fixes the data loss without scope creep.

Tests (tests/test_convert.py)

• test_serialize_bytes_encoding_fallback pins the contract: utf-8 passthrough is unchanged, and each of several non-UTF-8 byte strings serializes to its latin-1 string rather than None.
• test_json_non_utf8_bytes_preserved exercises the public to_json API on the shipped issue-463-example.pdf fixture, asserting both the top-level contents and the serialized data.Contents survive as strings.

Both fail on the current code with the value dropped to None/null and pass with the fix; tests/test_convert.py + tests/test_basics.py stay green (39 passed); black/isort/flake8 clean. CHANGELOG updated.