Fix heap out-of-bounds read in dimensional pcpatch WKB parsing

[HarshRajSinghania]

Summary

Fixes #387.

The dimensional pcpatch WKB parser trusted attacker-controlled per-dimension size fields and could read beyond the supplied WKB buffer in pc_bytes_deserialize(). The caller also advanced using the claimed size without tracking the remaining input length.

What changed

• Add an explicit bufsize parameter to pc_bytes_deserialize().
• Reject dimension records shorter than the 5-byte dimension header.
• Decode the size as signed int32_t and reject negative values.
• Reject any dimension whose claimed payload exceeds the remaining WKB buffer before allocation/copy.
• Pass the remaining WKB length through the dimensional parser and subtract each validated serialized dimension.
• Propagate pc_bytes_deserialize() failures and clean up partially constructed patches.
• Reject unexpected trailing bytes after all schema dimensions have been consumed.
• Add top-level WKB header-length validation before reading PCID/compression fields.
• Stop pc_patch_from_wkb() from continuing into extent/stat computation when a compression-specific parser returns NULL.

Regression coverage

Added CUnit coverage for:

• A 13-byte buffer claiming a 65536-byte dimension payload.
• A truncated 4-byte dimension header.
• A malformed dimensional patch reaching the public patch WKB dispatcher.

The malformed 65536-byte case is the regression for the demonstrated AddressSanitizer heap-buffer-overflow.

Validation

The branch is one commit ahead of master and the diff is limited to the parser/API changes and regression tests. I did not run the full CUnit/PostgreSQL build through this integration, so CI should be treated as the final build/test validation.

No RCE is claimed; the fix addresses the demonstrated out-of-bounds read and the associated oversized-allocation/DoS condition.