fix: Decode avro with negative count and no tags avro issue

[speeddragon]

Requesting https://arweave.net/raw/sfuxzQEEIFo5w6swrIPNjqUXCkaRm1BiuP5E3tmuNeU causes the header to be parsed incorrectly, leading to issues with non-UTF-8 binaries.

What's happening:

The Apache Avro array encoding has two forms:

1. Short-form: positive Count → Count items follow directly
2. Long-form: negative Count → abs(Count) items follow, but preceded by a zigzag-encoded byte size of the block (so readers can skip it)

The original decode_avro_tags only handled positive counts. The TXID sfuxzQEEIFo5w6swrIPNjqUXCkaRm1BiuP5E3tmuNeU is a real-world transaction whose tags were encoded using the long-form block (negative count), causing a crash/mismatch on deserialization.

Your current branch already has the fix (lines 594–598):

  decode_avro_tags(Binary, Count) when Count < 0 ->  
      {_ByteBlockSize, Rest} = decode_zigzag(Binary),
      decode_avro_tags(Rest, -Count);

This is standard Avro spec, not a custom encoding — it reads and discards the byte size, then recurses with abs(Count). The test at the bottom of the diff encodes the actual TX binary from that TXID and asserts the two expected tags (IPFS-Hash and Content-Type).

The fix is correct and complete. Nothing else needs to change — it's a missing clause in the existing Avro decoder, not a separate custom decoder.

---

This includes a second fix, on tx with no tags, like 8cjDy2khfMsc3hrvGp7PrLVYfD_4aYQxEILNSZ0Pv74