feat(tools): support Parquet and Arrow import with schema/auto modes

[gengziyand]

Summary

• Add Parquet and Arrow format support to the TsFile import tool (alongside existing CSV)
• Add auto mode (schema-less) import for all three formats
• Unify schema naming: id_columns → tag_columns, csv_columns → source_columns (backward compatible)
• Add parquet2tsfile.sh/.bat and arrow2tsfile.sh/.bat scripts
• Add CLI options: --format, --table_name, --time_precision, --separator

Changes

• New classes: ParquetSourceReader, ArrowSourceReader, AutoSchemaInferer, ImportExecutor, ImportSchema, ImportSchemaParser, SourceReader,
  SourceBatch, TabletBuilder, TimeConverter, ValueConverter
• Modified: TsFileTool.java (unified CLI entry point), CsvSourceReader.java (auto mode support)
• New dependencies: parquet-hadoop 1.14.4, hadoop-common 3.3.6, arrow-vector 15.0.2
• 189 automated tests across 11 test classes
• Updated README.md and README-zh.md

Test plan

• 189 unit and E2E tests passing (JDK 8)
• Smoke test on Linux JDK 17 (Docker) - 9 scenarios
• Smoke test on Windows JDK 8 and JDK 17 - 9 scenarios
• Data correctness verification - 39 checks (table name, TAG/FIELD, cross-format consistency)
• Boundary tests - 8 scenarios (tab separator, null values, SKIP/DEFAULT, error handling, fail_dir)

[CritasWang]

need check skip

[CritasWang]

In schema mode these readers drop skipped source columns from the SourceBatch, but TabletBuilder still indexes values by the original source_columns positions. With a schema like time, unused SKIP, value, getSchemaColumnNames() returns only [time, value], while TabletBuilder maps value to index 2, so batch.getValue(row, 2) throws and the import fails (or later columns shift incorrectly). This affects real Parquet/Arrow schemas using SKIP, and the current Parquet test only checks readBatch(), not the ImportExecutor path. Keep placeholder columns for skipped entries or change the builder mapping to use batch column names rather than schema positions.