test(sticktable): regression test for EntryUpdate.Marshal data encoding

[fionera]

What

Adds a regression test for EntryUpdate.Marshal's data encoding.

Why

EntryUpdate.Marshal used to encode the data section by calling MapData.Unmarshal instead of Marshal:

for _, data := range e.Data {
    n, err := data.Unmarshal(b[offset:]) // read, not write
    ...
}

That had two effects: the data section was written as zeros on the wire, and the source EntryUpdate's data was mutated to those zeros as a side effect. The code itself was fixed in #28 (write support) — this PR closes the test gap so it can't silently come back.

The existing TestMarshalUnmarshalMessages cannot catch this class of bug: it marshals entryUpdate, unmarshals into d, then compares *entryUpdate against d. Because the buggy Marshal mutates entryUpdate to zeros too, both sides match and the test passes even when the data is not encoded (verified by re-introducing the bug on the current tree — the existing test stays green, this new test fails with Marshal mutated source data to 0).

What this test does

• Marshals an EntryUpdate with a non-zero gpt0 value.
• Asserts the source data is not mutated by Marshal.
• Unmarshals into an independent struct and asserts the value round-trips.

No production code changes.

Context: I hit the underlying bug while building a peers sender on top of this library against v0.0.7; the fix is on master but not yet in a tagged release (v0.0.8 still ships data.Unmarshal). A release including it would be very welcome.