Cache ParsedVersion in FileMetaData to eliminate redundant created_by parsing

Start by reading FileMetaData, CorruptStatistics, and ColumnReadStoreImpl, then trace the listed VersionParser.parse(createdBy) call sites and the existing ParsedVersion overload in CorruptDeltaByteArrays. The work is done when FileMetaData exposes the cached parsed version, the requested overloads exist, and accessible call sites use the cached value without changing serialization behavior.

Summary

VersionParser.parse(createdBy) is called from 7 distinct production sites, all parsing the same constant string from FileMetaData.getCreatedBy(). Since FileMetaData is constructed once per file and already stores the createdBy string, it is the natural place to parse once and cache.

Problem

The created_by string is re-parsed into a ParsedVersion at every call site independently:

ColumnReadStoreImpl already parses createdBy into a ParsedVersion and stores it as a field — but does so R times (once per row group) because nobody upstream caches the parsed result.

Proposed Change

Add a cached ParsedVersion field to FileMetaData:

// In FileMetaData:
private final transient ParsedVersion writerVersion;

public FileMetaData(...) {
    ...
    this.createdBy = createdBy;
    this.writerVersion = parseVersion(createdBy);
    ...
}

public ParsedVersion getWriterVersion() {
    return writerVersion;
}

private static ParsedVersion parseVersion(String createdBy) {
    if (Strings.isNullOrEmpty(createdBy)) return null;
    try {
        return VersionParser.parse(createdBy);
    } catch (RuntimeException | VersionParseException e) {
        return null;
    }
}

Then add ParsedVersion-accepting overloads following the existing pattern established by CorruptDeltaByteArrays.requiresSequentialReads(ParsedVersion, Encoding):

• CorruptStatistics.shouldIgnoreStatistics(ParsedVersion, PrimitiveTypeName)
• ColumnReadStoreImpl constructor accepting ParsedVersion directly

Why This Approach

• Purely additive: new transient field + getter, no breaking changes
• Doesn't break serialization: field is transient
• Follows existing precedent: CorruptDeltaByteArrays already has ParsedVersion-based overloads used from ColumnReaderBase
• Enables incremental adoption: call sites can migrate to the cached version one at a time
• Unblocks #3607: PR #3607 can rebase onto this foundation cleanly using ParsedVersion-based APIs instead of threading a PARQUET-251-specific boolean

Scope

This issue covers:

1. Adding the ParsedVersion field and getter to FileMetaData
2. Adding shouldIgnoreStatistics(ParsedVersion, PrimitiveTypeName) overload to CorruptStatistics
3. Updating ColumnReadStoreImpl to accept ParsedVersion directly
4. Migrating existing call sites to use the cached version where FileMetaData is accessible

Context

Discussion: https://github.com/apache/parquet-java/pull/3607#issuecomment-5006760055
Related: #3601