[Pipe] Fix tablet type conversion for failed columns

[Caideyipi]

Description

Problem

A partial insert can leave a failed tablet measurement without a data type or value column. Pipe and Load type-conversion retries restore that failed measurement before schema validation, but their conversion overrides indexed the tablet arrays unconditionally. This could throw NullPointerException or ArrayIndexOutOfBoundsException instead of preserving the partial-insert result.

This is a follow-up to #17879, which made missing and null tablet columns valid partial-insert state across the write path.

Changes

• Validate the target type, measurement, source type, value column, and array bounds before Pipe tablet conversion.
• Reuse the same validation in the Load conversion override before logging or converting.
• Add regression coverage for missing columns, null failed measurements, and the Load conversion path.

Tests

• mvn spotless:apply -pl iotdb-core/datanode
• Targeted JUnit execution against the changed classes: OK (3 tests)
• The same tests against the pre-fix classes fail with the reproduced NPE/out-of-bounds behavior.

The regular single-module Maven test lifecycle was also attempted, but the local incremental build failed before test execution on unrelated generated fill/aggregation sources with missing symbols.

---

This PR has:

• been self-reviewed.
• added Javadocs for the non-trivial validation method.
• added comments explaining the intent where it is not obvious.
• added unit tests covering the new paths.

---

Key changed/added classes

• PipeConvertedInsertTabletStatement
• LoadConvertedInsertTabletStatement
• PipeConvertedInsertTabletStatementTest