Core: Fix AvroEncoderUtil header error printing literal %02X

[anxkhn]

## What

`AvroEncoderUtil.decode` validates the two Avro magic bytes with a Guava
`Preconditions.checkState`:

```java
Preconditions.checkState(
    header0 == MAGIC_BYTES[0] && header1 == MAGIC_BYTES[1],
    "Unrecognized header bytes: 0x%02X 0x%02X",
    header0,
    header1);

Iceberg's (relocated) Guava Preconditions formats messages with
Strings.lenientFormat, which only understands the %s placeholder - not
java.util.Formatter conversions like %02X. As a result the %02X specifiers
are emitted verbatim and the two byte arguments are appended as decimal in a
trailing bracket. A header mismatch therefore reports:

Unrecognized header bytes: 0x%02X 0x%02X [16, 32]

instead of the intended hex, which obscures the actual bytes that were read.

Fix

Build the message with String.format(Locale.ROOT, "Unrecognized header bytes: 0x%02X 0x%02X", header0, header1) and pass it as the single %s argument to
checkState. This mirrors the existing, correct handling of the identical message
in IcebergDecoder.decode
(core/src/main/java/org/apache/iceberg/data/avro/IcebergDecoder.java), and uses
Locale.ROOT per the project conventions for formatting.

With the fix, a mismatched header reports:

Unrecognized header bytes: 0x10 0x20

The thrown exception type is unchanged (checkState still throws
IllegalStateException); only the message text is corrected.

Testing

Added decodeRejectsUnrecognizedHeaderBytes to
core/src/test/java/org/apache/iceberg/avro/TestAvroEncoderUtil.java, asserting
that decoding a buffer with bad magic bytes throws IllegalStateException whose
message contains the rendered uppercase hex (and not the literal %02X). The test
fails before the fix (... was: "Unrecognized header bytes: 0x%02X 0x%02X [16, 32]") and passes after it.

• ./gradlew :iceberg-core:spotlessCheck - passes
• ./gradlew :iceberg-core:test --tests "org.apache.iceberg.avro.TestAvroEncoderUtil" - passes

---

AI Disclosure

• Model: Claude Opus 4.8
• Platform/Tool: opencode
• Human Oversight: partially reviewed
• Prompt Summary: Find and fix a wrong-diagnostic bug where a Guava Preconditions
  message used an unsupported %02X format specifier; apply the minimal fix
  mirroring the existing IcebergDecoder site and add a regression test.

Notes on the disclosure block (repo AGENTS.md mandates it for AI-authored PRs):
- `Human Oversight: partially reviewed` is honest - an independent automated
  reviewer verified the diff and re-proved the test red/green, and Anas gates the
  PR before it is opened, but no line-by-line human authoring happened. Anas can
  bump this to `fully reviewed` if he reads it end to end before opening.

---

## The diff (2 files, +12 / -3)

`core/src/main/java/org/apache/iceberg/avro/AvroEncoderUtil.java`
- add `import java.util.Locale;`
- replace the 3-arg `checkState(cond, "...0x%02X 0x%02X", header0, header1)` with
  `checkState(cond, "%s", String.format(Locale.ROOT, "Unrecognized header bytes:
  0x%02X 0x%02X", header0, header1))`

`core/src/test/java/org/apache/iceberg/avro/TestAvroEncoderUtil.java`
- add `assertThatThrownBy` + `org.junit.jupiter.api.Test` imports
- add `decodeRejectsUnrecognizedHeaderBytes()` asserting the message is
  `Unrecognized header bytes: 0x10 0x20`