avro: Bound and validate decoding against malformed input

[def-]

Hardens the mz-avro decoder against adversarial input (Avro bytes/schemas arrive from Kafka and an external registry, so a panic/OOM is an availability bug): bound per-block array/map lengths and object counts by remaining input, cap object-container block byte length, bound schema-parse/value-decode recursion, and fix two schema-resolution panics on unmatched named types.

Found by the cargo-fuzz suite (separate infra PR). Each fix has a regression test.

🤖 Generated with Claude Code

[martykulma]

Nice! Since we're introducing bounds that don't exist in the AVRO spec, this PR should also update the docs so customers are aware.

There's a subtle issue with remaining_input, see inline for the test case. Could we just drop the remaining input check and rely on the maximum element count * size_of::<Value>? We might want to drop the max element count in that case.

[martykulma]

subtle issue here for zero-byte elements

    #[mz_ore::test]
    fn valid_null_array_falsely_rejected() {
        // `remaining_input` assumes >=1 byte per element, false for null
        // (and other zero-byte) elements, so after the count VARINT, only
        // the 1-byte terminator remains. count >= 2 trips the check.
        use std::str::FromStr;

        use super::{AvroDeserializer, GeneralDeserializer};
        use crate::encode::encode_to_vec;
        use crate::types::Value;
        use crate::{Schema, ValueDecoder};

        let schema = Schema::from_str(r#"{"type": "array", "items": "null"}"#).unwrap();
        let value = Value::Array(vec![Value::Null, Value::Null]);
        let body = encode_to_vec(&value, &schema);

        let dsr = GeneralDeserializer {
            schema: schema.top_node(),
        };
        let mut reader: &[u8] = &body;
        let res = dsr.deserialize(&mut reader, ValueDecoder);
        assert!(
            res.is_ok(),
            "an encoder-produced array of nulls should round-trip, but got: {res:?}"
        );
    }

[martykulma]

nit: would you be opposed to calling this MAX_BLOCK_ELEMENTS - as I was reading this, I kept thinking this was bytes not element count.