Make Zod v4 the default output

[satvik007]

Summary

• Zod v4 is now the default schema output format
• Added zen.WithZodV3() option for backwards compatibility with snapshot-based workflows or incremental migration
• Updated README with migration guide and WithZodV3() usage example

Migration differences (v3 → v4)

• String format tags (email, http_url, ipv4, uuid4, md5) now use Zod v4 helpers: z.email(), z.httpUrl(), z.ipv4(), z.uuid({ version: "v4" }), z.hash("md5")
• ip / ip_addr emit z.union([z.ipv4(), z.ipv6()])
• Embedded anonymous structs now expand via .shape spreads instead of .merge(...)
• Enum-like map keys emit z.partialRecord(...)

Test plan

• Existing tests pass with Zod v4 as default
• WithZodV3() option produces previous v3 output
• README examples are accurate

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request transitions the default output to Zod v4 schemas, adding a WithZodV3 option for legacy support. Key updates include refactoring struct conversion to use shape spreads, adopting specialized Zod v4 string format helpers, and implementing z.partialRecord for enum-keyed maps. Review feedback points out a logic error in struct conversion where named fields are incorrectly overridden by embedded spreads due to improper ordering; spreads should be emitted first to ensure outer struct fields take precedence. Additionally, a correction was suggested for the datetime format helper to use z.datetime() instead of z.iso.datetime() to align with Zod v4 conventions.

[gemini-code-assist]

Similar to getStructShape, there's a field ordering issue here for Zod v4 schemas. Named fields are written to the output before embedded fields are spread, which will cause properties from embedded structs to incorrectly override properties from the outer struct if they share the same name. The spreads should come first to ensure correct overriding behavior.

  merges := []string{}
  embeddedFields := []string{}
  namedFields := []string{}

  fields := input.NumField()
  for i := 0; i < fields; i++ {
    field := input.Field(i)
    optional := isOptional(field)
    nullable := isNullable(field)

    if field.Anonymous {
      if c.zodV3 {
        line, shouldMerge := c.convertEmbeddedFieldMerge(field, indent+1)
        if shouldMerge {
          merges = append(merges, line)
        } else {
          output.WriteString(line)
        }
      } else {
        embeddedFields = append(embeddedFields, c.convertEmbeddedFieldSpread(field, indent+1))
      }
    } else {
      if c.zodV3 {
        output.WriteString(c.convertNamedField(field, indent+1, optional, nullable))
      } else {
        namedFields = append(namedFields, c.convertNamedField(field, indent+1, optional, nullable))
      }
    }
  }

  if !c.zodV3 {
    for _, line := range embeddedFields {
      output.WriteString(line)
    }
    for _, line := range namedFields {
      output.WriteString(line)
    }
  }

[gemini-code-assist]

It seems there might be a typo here. The Zod v4 proposal for string formats suggests z.datetime() for datetime strings, not z.iso.datetime(). Was z.datetime() intended?

Also, it would be good to add a test case for the new datetime behavior in TestZodV4Defaults.

Suggested change

	chunks = append(chunks, stringSchemaChunk{kind: "format", v4Base: "z.iso.datetime()", legacyChain: ".datetime()"})
	chunks = append(chunks, stringSchemaChunk{kind: "format", v4Base: "z.datetime()", legacyChain: ".datetime()"})