Revision history for PostgreSQL extension tdigest.

1.4.7
    - Keep percentile interpolation monotone for closely spaced means and ranks.
    - Use consistent inverse-percentile ranks at centroid boundaries with large counts.
    - Disable FP contraction for LLVM bitcode as well as for native code.
    - Reject text input that underflows to zero, while accepting nonzero subnormals.
    - Improve uninstall to cleanup this extension's control and SQL files.
    - Mark scalar functions operating on a tdigest value as parallel safe, to
      allow parallel queries in more cases.
    - Avoid division by zero when rebalancing and compacting digests (benign with IEEE-754).
    - Preserve reusable tdigest states during finalization (addresses `FINALFUNC_MODIFY = READ_ONLY`).
    - Improve error messages for arrays containing NULL values.
    - Fix duplicate keys in tdigest_to_json() output.
    - Reject NULL compression in the aggregate transition functions.
    - Reduce memory usage (grow/shrink aggregate state, eagerly free copies, read/write float8 arrays directly).
    - Corrections and clarifications of assorted comments and docs.
    - Various cleanups of stale boilerplate in the Makefile.

1.4.6
    - Disable FP contraction to make tests pass on arm64
    - Fix builds on 32-bit systems by defining strtoi64
    - Fix validation of NULL / empty in several functions
    - Clamp results to valid range to handle rounding errors
    - Make possibly-long loops interruptible
    - Make centroid sort more resilient to invalid data

1.4.5
    - Fix multiple memory-safety issues (OOB reads/writes, SIGSEGV crashes)
    - Fix multiple issues with float precision (NaN/infinity results)
    - Fix calculation of low/high thresholds in trimmed aggregates
    - Fix trimmed sum/avg aggregates to sort the centroids
    - Make sure processing does not generate digests with NaN/infinity float values
    - Make sure various int64 values do not overflow
    - Harden validation of function parameters to reject NULL/NaN/infinity values
    - Use overflow-resilient and/or numerically stable formulas in multiple functions
    - Improve test coverage by adding new regression tests
    - Speed up regression tests by using a parallel schedule
    - Add tdigest_is_valid() to check validity of existing t-digest values
    - Mitigate incorrect alignment of t-digest values.

1.4.4
    - Add missing parts of automated release workflow.
    - Make regression tests pass on Postgres 19 branch
    - Fix various issues when parsing/validating input data
    - Fix combining t-digests with different parameters

1.4.3
    - PGXN packaging (META etc.).
    - Automating releases to GitHub / PGXN
