fix(canonical): prevent field names being substituted with type names in CanonicalSchema

[jwilliams-vc]

Fixes #307.

Problem

CanonicalSchema() produces two different strings from the same schema non-deterministically (~14% vs ~86% split) when a field name and a record type share the same local name within the same inherited namespace.

Minimal reproduction:

const schema = `{
  "type": "record", "name": "Parent", "namespace": "com.example",
  "fields": [{
    "name": "items",
    "type": ["null", {"type": "array", "items": {"type": "record", "name": "items", "fields": [{"name": "id", "type": "int"}]}}],
    "default": null
  }]
}`

seen := map[string]int{}
for i := 0; i < 1000; i++ {
    codec, _ := goavro.NewCodec(schema)
    seen[codec.CanonicalSchema()]++
}
// len(seen) == 2  (should always be 1)

Correct (~86%): ..."fields":[{"name":"items","type":...
Wrong (~14%): ..."fields":[{"name":"com.example.items","type":...

Field names must never be namespace-qualified per the Avro PCF spec.

Root cause

pcfObject iterates over the JSON map (random Go order). If the "type" value is visited before the "name" value, the inner record type "items" gets registered in typeLookup as "com.example.items". When pcfString later processes the field's "name" value, it finds "items" in typeLookup and substitutes the fully-qualified type name — incorrectly namespace-qualifying the field name.

Fix

typeLookup exists to resolve type references, not identifiers. A "name" value is either a type's own name (already namespace-qualified by the block above) or a field/symbol name that must be emitted verbatim — neither should go through typeLookup resolution.

Rather than special-casing the string value inside the k == "name" branch, we pass nil as typeLookup when recursing over the value of a "name" key. This expresses the constraint at the call site: the entire subtree below a "name" key sees no type lookup table. nil map reads are safe in Go, so pcfString returns the value verbatim as intended.