[#2601] feat(spark): Overlapping decompression for shuffle read

[zuston]

What changes were proposed in this pull request?

This PR is to introduce the overlapping decompression for the shuffle reading for the better shuffle speed.

Why are the changes needed?

for #2601

When applying the #2598 into the benchmark of terasort 100g, I found some bottleneck for the decompression time in the read phase.

Based on a 100 GB Terasort benchmark, the results are impressive, reducing shuffle read time by 50% after applying this PR.

Does this PR introduce any user-facing change?

Yes

How was this patch tested?

1. Unit tests.

[github-actions]

Test Results

 3 105 files  +15   3 105 suites  +15   6h 49m 38s ⏱️ +9s
 1 200 tests + 2   1 199 ✅ + 3   1 💤 ±0  0 ❌ ±0 
15 196 runs  +30  15 181 ✅ +32  15 💤 ±0  0 ❌ ±0 

Results for commit deb3a6c. ± Comparison against base commit 2a32171.

♻️ This comment has been updated with latest results.

[jerqi]

Spark is ok. But mr may need order.

[zuston]

Spark is ok. But mr may need order.

Thanks for sharing this, but the impl has ensured the order. BTW now this PR is only valid in spark