fix(spec): use canonical fixed[N] type string in PrimitiveType Display

[laskoviymishka]

Which issue does this PR close?

• N/A — trivial spec-conformance fix. Happy to file a tracking issue if preferred.

What changes are included in this PR?

PrimitiveType's Display impl emitted fixed(N) (parentheses), diverging from the
canonical Iceberg type string fixed[N] (square brackets) used everywhere else:

• the serde path — serialize_fixed already emits fixed[{value}] and
  deserialize_fixed parses fixed[;
• the reference implementations — iceberg-java (Types.FixedType.toString()) and
  iceberg-go (FixedType.String()) both use fixed[N], as does PyIceberg.

Fixed was the only Display arm inconsistent with its own serialized form
(Decimal, for comparison, matches: decimal(p, s) in both). The likely origin was
a copy of the decimal(...) paren style.

This PR:

• changes the Display arm to fixed[{size}];
• adds test_primitive_type_display, a table-driven test over every PrimitiveType
  variant that pins all three directions — Display, Serialize, and
  Deserialize — to the same canonical string, so this class of drift is caught in
  future.

No behavioral impact on any wire/interop path: the serde (de)serialization was already
correct, and Display has no FromStr counterpart — its output is consumed only in
diagnostic/error strings. This is purely a bug fix, not a format change to persisted data.

Are these changes tested?

Yes. New unit test test_primitive_type_display in
crates/iceberg/src/spec/datatypes.rs asserts, for every primitive variant (including
Fixed(4), the Fixed(0) edge case, and Decimal boundary values):

1. Display emits the canonical string (fixed[4], not fixed(4));
2. Serialize produces the same string;
3. Deserialize round-trips it back to the original type.

The pre-existing fixed(N) bug was invisible precisely because the serde path was
already correct — no round-trip test exercised Display directly. This test closes
that gap. Verified locally: cargo test -p iceberg spec::datatypes (12 pass),
cargo fmt --check, and cargo clippy all clean.

[huan233usc]

LGTM, thanks for the fix