Simplify Decoder API to use byte slices

[tamirms]

Summary

• Replace the io.Reader-based Decoder with a new Decoder that operates directly on byte slices
• Add Decode(v DecoderFrom) convenience method for decoder reuse with Reset()

Key API Changes

Before	After
Unmarshal(io.Reader, v)	Unmarshal([]byte, v)
NewDecoder(io.Reader)	NewDecoder([]byte)

New Decoder Reuse Pattern

decoder := xdr.NewDecoder(nil)

// Decode multiple values, reusing the decoder
decoder.Reset(data1)
n, err := decoder.Decode(&value1)

decoder.Reset(data2)
n, err = decoder.Decode(&value2)

Benchmarks show ~10% improvement with decoder reuse.

Breaking Changes

• Unmarshal now takes []byte instead of io.Reader
• NewDecoder now takes []byte instead of io.Reader
• Minimum Go version is now 1.22

Alternatively, I could introduce these changes in an xdr4 package to avoid breaking changes. However, I think we are the only users of this library and I preferred to remove existing code to make the codebase easier to maintain. But I am open to reconsidering this decision.

[leighmcculloch]

Before	After
Unmarshal(io.Reader, v)	Unmarshal([]byte, v)
NewDecoder(io.Reader)	NewDecoder([]byte)

Would this mean that it's no longer possible to stream decode some XDR types? I'm thinking of a bucket which is a stream of framed bucket entries. How would I decode bucket entries with it being a byte slice? Does that mean the entire bucket has to be loaded into memory in a byte slice? There may be some cases where it would be preferred to still be able to use a reader.

[leighmcculloch]

Benchmarks show ~10% improvement with decoder reuse.

Do we understand what's causing this improvement? I'm wondering if this improvement a result of removing the decoder interface, or if it's to do with buffering. Because we can do buffered reads using the bufio types and retain the reader interface.