iceberg: add case_sensitive_columns option

[Jeffail]

Summary

• Adds a top-level case_sensitive_columns boolean to the iceberg output (default true, preserving released behaviour). When set to false, column-name matching follows iceberg's recommended case-insensitive convention end-to-end.
• The flag is plumbed through every site that compares iceberg field names: shredder lookup, schema_metadata traversal, partition spec column references, struct inference (auto-inferred and metadata-driven), top-level table-creation duplicate detection, sink dedup keys for cross-message new fields, and the caseSensitive argument to iceberg-go's UpdateSchema. Ambiguous case-only duplicates in input records are rejected at shredding and creation.
• Resolves a customer-reported issue where messages with upper-case keys against a table with lower-case columns triggered spurious schema evolution rather than routing into the existing columns.

Test plan

• task fmt
• task lint (0 issues)
• task test:unit
• New unit tests across the shredder, type inferrer/resolver, partition-spec parser, router, and writer sinks — covering both the case-insensitive path and the case-sensitive default
• Integration tests (internal/impl/iceberg/integration):
  • CaseInsensitiveColumnMatching — pre-created lowercase table, uppercase-keyed messages, no schema evolution, data lands in the right columns
  • CaseInsensitiveCreateTimeDuplicate — record with id and ID rejected at create time
  • CaseInsensitiveNewFieldDedupAcrossBatch — two messages with case-only-different new fields produce exactly one new column
  • CaseInsensitivePartitionSpecCreate — partition spec with uppercase column reference resolves against schema inferred with lowercase keys

[claude]

Commits
LGTM

Review
Adds an end-to-end case_sensitive_columns option for the iceberg output, plumbed through the shredder, schema_metadata resolution, partition spec parser, struct inference, table creation, sink dedup, and catalogx UpdateSchema. Defaults to true for backwards compatibility, with explicit ambiguity detection at every layer where case-only duplicates could otherwise cause silent data loss or schema corruption. Unit tests cover each touchpoint and four integration tests exercise the column-matching path, create-time duplicate rejection, cross-message new-field dedup, and case-insensitive partition spec parsing.

LGTM

[claude]

Commits
LGTM

Review
Comprehensive plumbing of the new case_sensitive_columns flag through every iceberg comparison site (shredder, schema_metadata traversal, partition spec parser, type inference/resolver, sink dedup, UpdateSchema). Default true preserves released behaviour. Case-only ambiguity is rejected at the right points. Unit and integration coverage exercise both modes.

LGTM

[claude]

Commits
LGTM

Review
The PR threads a new case_sensitive_columns config flag (default true for backward compatibility) through the iceberg output's shredder, type inferrer, type resolver, partition spec parser, sinks, and catalogx.UpdateSchema. Case-only duplicate detection is added at table creation, struct inference, and shredding, and new-field dedup folds case in case-insensitive mode. Unit and integration coverage is comprehensive across each touchpoint.

LGTM