[VL] Story: Improve on data ingestion workloads

[zhouyuan]

Description

Currently Gluten supports reading on Iceberg/Hudi/Delta tables natively. However on the write path Gluten does not support these and will use JVM Spark code.

On Iceberg side, Velox community already have a working patch with write support(non-bucketed): facebookincubator/velox#10996

Note this heavily depends on the C++ write support of Iceberg/Hudi/Delta. A unified design will beneficial for all backends.

[zhouyuan]

On Iceberg support #8953

Read:

• Support reading CoW - done

• Support reading MOR - basic support ready, w/o equality delete

• Equality delete support

Design: facebookincubator/velox#8748
Patch: facebookincubator/velox#11088
Gluten patch: #8056
Positional delete: facebookincubator/velox#12590

Writing - fallback with JVM code, we need a design on this

• CoW vs MoR: this should be handled in Spark iceberg connector
• Non-bucket: feat(iceberg): Add support for writing iceberg tables (to be closed) facebookincubator/velox#10996
• Bucket: pending on new patch from velox community
• Write data from gluten, metadata write from JVM spark

Unit tests coverage:
#8932

Performance report:
Iceberg perf: Java vs. native

Documents update:
TBD