fix(attachments,usage): validate content behind a BOM, judge by the live parser table

Review follow-ups.

A BOM told the sniff which encoding to read, but was also taken as the
verdict: three prepended bytes let any binary through, including as
notes.txt. A BOM now only selects the test — UTF-8 falls through to the
byte rules on the remainder, UTF-16/32 decode and judge the characters
(NUL, unprintable, or replacement chars from bytes the decoder could not
read). Real Notepad-Unicode text still passes, mp4-behind-a-BOM does not,
in either language.

The gate treated the full parser table as a given, but without docling the
fallback extractor has no .tif/.tiff/.bmp/.webp/.vtt/.xml handler, so those
suffixes skipped the content check and reached the plain-text fallthrough —
the original bug, one install away. The worker now passes the keys of the
extractor it actually built, making the second gate stricter than the
route's static one rather than a copy of it.

Cache bins: extraction coerced a missing value to 0 and only non-zero bins
were recorded, so a provider reporting cached_tokens=0 persisted as NULL —
indistinguishable from "not reported", and OpenAI reports exactly that on
every uncached request. Bins are now carried as Optional and recorded when
not None, which is what the nullable columns and the NULL-means-unknown
comment already assumed. Anthropic's cache_read/cache_creation bins had the
same shape and are fixed alongside; the int-or-None coercion is shared in
llm/base.py.

tests/test_upload_limits.py

@@ -121,6 +121,49 @@ def test_enforce_parseable_attachment_accepts_bom_marked_unicode_text(
     enforce_parseable_attachment(path, "notes.txt")
 
 
+@pytest.mark.parametrize(
+    "bom",
+    [codecs.BOM_UTF8, codecs.BOM_UTF16_LE, codecs.BOM_UTF16_BE, codecs.BOM_UTF32_BE],
+)
+def test_enforce_parseable_attachment_rejects_binary_behind_a_bom(bom, tmp_path):
+    """A BOM says which encoding to read, not that the content is text.
+
+    Otherwise three prepended bytes buy any binary a pass.
+    """
+    from application.upload_limits import (
+        enforce_parseable_attachment,
+        UnsupportedUploadTypeError,
+    )
+
+    path = tmp_path / "notes.txt"
+    path.write_bytes(bom + MP4_HEADER + bytes(range(256)) * 8)
+
+    with pytest.raises(UnsupportedUploadTypeError):
+        enforce_parseable_attachment(path, "notes.txt")
+
+
+def test_enforce_parseable_attachment_uses_the_extractor_it_is_given(tmp_path):
+    """The worker holds the live parser table; a trimmed install must not admit on trust.
+
+    Without docling the fallback extractor has no .webp handler, so a .webp
+    would otherwise skip the content check and be read as plain text.
+    """
+    from application.upload_limits import (
+        enforce_parseable_attachment,
+        UnsupportedUploadTypeError,
+    )
+
+    path = tmp_path / "scan.webp"
+    path.write_bytes(b"RIFF\x00\x00\x00\x00WEBPVP8 " + bytes(range(256)))
+
+    # Default list: .webp is parser-backed, admitted on its name.
+    enforce_parseable_attachment(path, "scan.webp")
+
+    # The extractor actually loaded has no .webp parser.
+    with pytest.raises(UnsupportedUploadTypeError):
+        enforce_parseable_attachment(path, "scan.webp", {".pdf", ".docx"})
+
+
 @pytest.mark.parametrize(
     "filename",
     [
@@ -177,6 +220,10 @@ def test_enforce_parseable_attachment_accepts_text_without_a_parser(filename, tm
         (b"\x1b[31mred log line\x1b[0m\n", True),
         (codecs.BOM_UTF16_LE + "hi\n".encode("utf-16-le"), True),
         (codecs.BOM_UTF8 + b"hi\n", True),
+        (codecs.BOM_UTF32_LE + "hi\n".encode("utf-32-le"), True),
+        # A BOM in front of binary is still binary.
+        (codecs.BOM_UTF8 + b"\x00\x01\x02", False),
+        (codecs.BOM_UTF16_LE + MP4_HEADER, False),
         (b"text\x00with nul", False),
         (bytes(range(32)) * 4, False),
         (b"\x7f\x7f\x7f\x7f" + b"a" * 16, False),