maintenance: preserve metric string values

[zqr10159]

Summary

• remove the arbitrary 32,700-byte per-cell truncation from the shared metrics representation
• let Arrow VarCharVector.setSafe grow its variable-width data buffer as designed
• decode metric cells explicitly as UTF-8
• add exact-fidelity contracts around the previous boundary and through Arrow IPC

Boundary analysis

The previous 32,768-byte comment described Arrow initial buffer capacity, not a logical cell limit. Apache Arrow 18.1 documents that BaseVariableWidthVector.setSafe(int, byte[]) handles values beyond existing capacity and that the data buffer is reallocated as needed: https://arrow.apache.org/docs/18.1/java/reference/org.apache.arrow.vector/org/apache/arrow/vector/BaseVariableWidthVector.html#setSafe(int,byte%5B%5D)

CollectRep.MetricsData is the common in-memory representation and is serialized as Arrow IPC for Kafka and Redis paths. No universal 32,700-byte database or protocol limit exists at this layer. Deployment-specific transport limits apply to the complete serialized message, not silently to one metric cell; any future transport-specific limit must be enforced and reported at that output boundary.

Regression evidence

Before removing the truncation, the new contracts failed five times: values immediately above 32,700 bytes and a 100,000-byte value lost data in memory, and the 100,000-byte value was also truncated after Arrow IPC serialization and deserialization.

The final contracts cover ASCII, a three-byte code point, and a four-byte emoji immediately before, at, and after the old boundary, plus a 100,000-byte ASCII value. They require exact content and UTF-8 byte-length preservation. The IPC contract requires exact preservation after serialization and deserialization.

Validation

• ./mvnw -pl hertzbeat-common-core -Dtest=CollectRepTest,KafkaMetricsDataSerializerTest test -DskipITs -Dsurefire.failIfNoSpecifiedTests=false -DfailIfNoTests=false — 18 tests passed
• ./mvnw -pl hertzbeat-common-core test -DskipITs — 595 tests passed
• Checkstyle, no hard-coded CJK scan, and both Git diff whitespace checks passed

AI assistance: used for draft implementation and test iteration.
Human validation: the previous truncation failed exact-fidelity regressions; the focused Arrow/IPC contracts and complete common-core suite pass after removing it.
Risk notes: very large metric cells consume proportional memory and contribute to whole-message transport limits. Those limits are deployment-specific and must fail observably at the relevant transport boundary; this shared layer no longer corrupts values silently.

[zqr10159]

Author remediation update:

The arbitrary 32,700-byte truncation has been removed. Arrow VarChar storage grows through setSafe, preserving full UTF-8 values instead of silently changing metric data. The regression suite covers ASCII, CJK, emoji, exact boundaries, over-boundary input, a 100 KB value, and Arrow IPC round-trip behavior.

Focused tests (18) and the complete common-core suite (595 tests) passed; all current GitHub checks are green. The data-integrity blocker is resolved; maintainer review is still required.