fix(query): preserve scalar bloom hashes in replace into

[dantengsky]

I hereby agree to the terms of the CLA available at: https://docs.databend.com/dev/policies/cla/

Summary

Distributed REPLACE INTO can panic when an external Parquet source projects a constant conflict
key:

panicked at deletion_accumulator.rs:45:25:
assertion `left == right` failed
  left: 3
 right: 4

The SELECT-list constant remains a Scalar in a locally processed source block, but remote exchange
materializes it as a Column. The previous Bloom hashing code silently skipped the Scalar, so deletion
information for the same target block contained three Bloom hash columns from a local block and four
from a remote block, triggering the assertion.

This PR materializes every selected Bloom key as a full Column before hashing. This prevents the
panic and keeps source Bloom hashes aligned with their target columns.

Changes

• Replace filter_map(Value::as_column) with full-column materialization before Bloom hashing.
• Add a unit regression covering Scalar and Column representations in one deletion accumulator.
• Add a cluster-only regression with an external FROM 'fs:///...' Parquet source, a SELECT-list
  constant, an explicit three-node assertion, stable source/target partition placement, and an
  actual conflicting row replacement.

Verification

• The exact cluster regression on unfixed debug build v1.2.925-nightly-1f0ac2a483 reproduces the
  left: 3, right: 4 panic.
• With this fix on the same debug version, the exact cluster regression passes. The target row count
  is 23, and the conflicting row is replaced with the source value.
• cargo fmt --all -- --check
• git diff --check

Tests

• Unit Test
• Logic Test
• Benchmark Test
• No Test - Explain why

Type of change

• Bug Fix (non-breaking change which fixes an issue)
• New Feature (non-breaking change which adds functionality)
• Breaking Change (fix or feature that could cause existing functionality not to work as expected)
• Documentation Update
• Refactoring
• Performance Improvement
• Other (please describe):

---

This change is

[dantengsky]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Nice work!

Reviewed commit: 4450123095

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".