Fix id write endianness and sub-width signed sign-extension

[gusanthon]

Summary

Two independent correctness fixes for byte-aligned reads/writes, each with
regression tests. They're unrelated bugs that happened to surface together
while implementing a binary codec - happy to split into two PRs if you'd
prefer to review/merge them separately.

---

Patch A - enum id written with the wrong endianness

File: deku-derive/src/macros/deku_write.rs

On the id_pat write path, field_endian was derived from id_endian
only, ignoring both a hardcoded endian = "big" and a bound ctx endian.
As a result an enum that declares endian = "big" would still write its
id in little-endian, so the written bytes didn't match the equivalent
flat big-endian field (and didn't round-trip against the read side).

Fix: mirror the read-side precedence already used in gen_id_args
(id_endian -> enum endian) for the id_pat write path, so the write side
resolves endianness the same way the read side does.

Tests (tests/test_enum.rs):

• an id_pat enum with endian = "big" writes the same bytes as the
  equivalent flat big-endian field
• a signed 24-bit id_pat enum round-trips negative sentinels on both LE
  and BE

---

Patch B - sub-width signed reads not sign-extended

File: src/impls/primitive.rs

The (Endian, ByteSize, Order) DekuReader impl for signed types
(ImplDekuReadSignExtend macro) zero-padded the high bytes of a sub-width
byte-aligned read without propagating the sign bit. E.g. a 3-byte i32
read of -100 returned +16776092 instead of -100.

Fix: apply the same shift-based sign-extension already used in the
BitSize path, so the ByteSize path sign-extends consistently.

Tests (src/impls/primitive.rs): cover the ByteSize sign-extend path
directly for i32 and i16 sub-width reads on both endians.

[wcampbell0x2a]

The following would panic with this patch:

#[derive(Debug, DekuRead)]
struct Dynamic {
    n: u8,
    #[deku(bytes = "*n as usize")]
    a: i32,
}
#[test]
fn dynamic_zero() {
    // n = 0 -> ByteSize(0) at runtime
    let mut c = Cursor::new([0x00u8, 0x01, 0x02]);
    println!("{:?}", Dynamic::from_reader((&mut c, 0)));
}


}

[wcampbell0x2a]

Requesting changes, see previous comment.

[gusanthon]

Thanks for the review, @wcampbell0x2a, and good catch. Fixed in a4debfa by skipping the sign-extension when the size is 0. Your snippet gives Ok((8, Dynamic { n: 0, a: 0 })) again, same as master. Added it as a test in
tests/test_regression.rs. Also rebased onto master, so no more merge commits. Let me know if there's anything else that needs addressing.