Website: Add blog post for 25.0.0

[raulcd]

Release blog post for 25.0.0.

I am slightly late this time so Release Notes are already available here:
https://arrow.apache.org/release/25.0.0.html

Issues on the milestone are here:
https://github.com/apache/arrow/milestone/74?closed=1

[raulcd]

@rok @pitrou can you help me with the C++ / Parquet notes?

[raulcd]

@kevingurney @sgilmore10 any Matlab note for the blog post for 25.0.0?

[kevingurney]

@raulcd - thanks for checking! I don't think there is anything worth specifically noting for 25.0.0.

[raulcd]

@AlenkaF can you help with the release notes for Python?

[raulcd]

@thisisnic @jonkeane could you help with the R notes?

[github-actions]

Preview URL: https://raulcd.github.io/arrow-site

If the preview URL doesn't work, you may have forgotten to configure your fork repository for preview.
See https://github.com/apache/arrow-site/blob/main/README.md#forks for instructions on how to configure.

[rok]

@raulcd I've generated and manually reviewed these. Perhaps @pitrou could say something about the security push that was done for C++.

We should also mention CI caching improvements somewhere!

[rok]

Suggested change

	Sorting, select-k and ranking APIs can now specify null placement independently for each `SortKey`. The older global null-placement options are deprecated [GH-38558](https://github.com/apache/arrow/issues/38558).
	A new `hypot` compute function calculates Euclidean norms without the avoidable overflow of a naive `sqrt(x*x + y*y)` implementation [GH-50197](https://github.com/apache/arrow/issues/50197). Comparison kernels now support BinaryView and StringView inputs [GH-46856](https://github.com/apache/arrow/issues/46856).
	Several correctness issues were also fixed. Ranking now treats NaNs and nulls as distinct values when resolving ties [GH-45193](https://github.com/apache/arrow/issues/45193), `count` now accounts for logical nulls in run-end encoded arrays [GH-49888](https://github.com/apache/arrow/issues/49888), and sorting temporal columns in a table no longer crashes [GH-47252](https://github.com/apache/arrow/issues/47252).

[rok]

Suggested change

	`Table` values can now be converted to an Arrow `Tensor` [GH-40062](https://github.com/apache/arrow/issues/40062).
	Deserialization of canonical tensor extension metadata is now stricter. In particular, fixed-shape tensor metadata validates shapes, permutations, dimension names and consistency with the storage type; related variable-shape tensor metadata is validated more thoroughly as well [GH-49716](https://github.com/apache/arrow/issues/49716).

[rok]

Suggested change

	Parquet C++ can now write ListView and LargeListView arrays. When the Arrow schema is stored in the file, readers can restore the original view type; otherwise they read back as List or LargeList [GH-38849](https://github.com/apache/arrow/issues/38849).
	Split-block Bloom filters can now be folded before serialization, allowing the writer to start with a conservative size and automatically reduce on-disk size to meet the target false-positive probability
	[GH-50007](https://github.com/apache/arrow/issues/50007). Bloom-filter probes were also made branchless and SIMD-accelerated, including a runtime-dispatched AVX2 implementation [GH-50026](https://github.com/apache/arrow/issues/50026).
	The bundled Thrift compiler was upgraded from 0.21.0 to 0.23.0 [GH-50267](https://github.com/apache/arrow/issues/50267) and `parquet.thrift` was synced with Parquet format 2.13.0 [GH-50265](https://github.com/apache/arrow/issues/50265).
	Parquet decoding was hardened against malformed inputs and excessive temporary allocations in DELTA_BINARY_PACKED, DELTA_LENGTH_BYTE_ARRAY, DELTA_BYTE_ARRAY and BYTE_STREAM_SPLIT data ([GH-49805](https://github.com/apache/arrow/issues/49805), [GH-49837](https://github.com/apache/arrow/issues/49837), [GH-49959](https://github.com/apache/arrow/issues/49959)). Timestamp coercion now detects integer overflow instead of silently writing corrupt values [GH-47657](https://github.com/apache/arrow/issues/47657).
	Statistics handling received several correctness fixes. Min/max statistics are ignored when an unknown column order makes them unsafe to interpret [GH-50156] (https://github.com/apache/arrow/issues/50156), and all-infinity floating-point columns now produce exact infinity statistics rather than finite limits [GH-50182](https://github.com/apache/arrow/issues/50182).

[rok]

Suggested change

	IPC reading was hardened substantially. Readers now reject short buffers and inconsistent message sizes, and sparse tensor metadata receives stricter shape, buffer-size, count and overflow validation ([GH-49896](https://github.com/apache/arrow/issues/49896), [GH-50240](https://github.com/apache/arrow/issues/50240), [GH-50054](https://github.com/apache/arrow/issues/50054), [GH-50077](https://github.com/apache/arrow/issues/50077), [GH-50161](https://github.com/apache/arrow/issues/50161), [GH-50304](https://github.com/apache/arrow/issues/50304)).
	Floating-point equality can now use a configurable ULP distance through `EqualOptions` [GH-48408](https://github.com/apache/arrow/issues/48408).
	The filesystem factory interface can accept key-value options separately from the URI. S3 factory options can therefore carry credentials and other user-specific settings without embedding them in a reusable URI [GH-46369](https://github.com/apache/arrow/issues/46369).
	On AArch64, runtime SIMD dispatch now supports SVE for bit-unpacking operations [GH-47769](https://github.com/apache/arrow/issues/47769).

[lidavidm]

Suggested change

	## Arrow Flight RPC Notes
	## Arrow Flight RPC Notes
	The Flight SQL protocol was amended to let servers explicitly inform clients whether a prepared statement contains a result set or not ([GH-49497](https://github.com/apache/arrow/issues/49497)).
	Progress was made on the ODBC driver for Flight SQL, but we are not yet distributing packages for end users at this time.

[lidavidm]

@amoeba @justing-bq I see the GitHub release does contain an MSI installer; is that actually ready to use?

[amoeba]

It is. Not all the DLLs are signed (yet) so it's usable with the caveat that some users may need to work around Defender. There are no docs yet either but I think this is fine as a release note:

This release includes a new, official Windows installer for the ODBC driver for FlightSQL. Installers for Linux and macOS are still in progress. Note: The installer is signed by the Apache Software Foundation but some DLLs inside the installation directory are not signed and may need to be manually unflagged to use the driver.

Thanks for pointing this out.