fix(schema): a SchemaField's unread octets are lost through from_data, so a direct SchemaField rebuilds zero-filled #1380

Split from #1292 after the #1374 cross-review. #1374 makes a directly declared SchemaField pack its whole declared span, so the parse → pack round trip is byte-exact: 0302eeee04 rebuilds as 0302eeee04. The parse → from_data round trip still loses the unread octets. Out2.from_dict(o.to_dict()).pack() gives 0302000004. The length is consistent, but the octets the inner schema did not read are zero-filled.

The cause is that Protocol.from_data (protocols/protocol.py:799-847) runs Data.from_dict → _make_data → __init__ → make. That path builds a fresh schema that was never unpacked, so there are no unread octets to re-emit, and the data model has no slot that carries them.

Scope: none of the 446 shipped schema classes declares a SchemaField directly or behind a static ConditionalField. Every library SchemaField sits behind a SwitchField or is a ListField item. This affects user-defined schemas only.

Fix: carry the unread remainder through the data model, e.g. a remainder key in to_dict(), and re-emit it on build, as #1344 did for IPv4 option padding and #1348 for TCP. Then update #1374's test_schema_field_keeps_declared_width, which currently pins 0302000004.

Blocked on #1374 (same schema.py).