chore: prepare changelog for 2.6.4 release (#21729)

- Add [2.6.4] release section with all changes from the skipped 2.6.2/2.6.3
- Note supply chain security compromise (see #21691) as reason for skipped versions
- Includes: Neptune logger removal, LitLogger bump, val_check_interval fix,
  pkg-resources fix, FSDP mixed precision fix, device_mesh type hint fix,
  RichModelSummary fix, SimpleProfiler math.fsum precision fix

Author: bhimrazy

src/lightning/fabric/CHANGELOG.md

@@ -22,14 +22,15 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 -
 
---
+---
+
+## [2.6.4] - 2026-05-20
 
-## [2.6.2] - 2026-03-19
+> Versions 2.6.2 and 2.6.3 were skipped due to a supply chain security compromise. See [#21691](https://github.com/Lightning-AI/pytorch-lightning/issues/21691) for details.
 
 ### Fixed
 
 - Fixed FSDP mixed precision (`bf16-mixed`, `16-mixed`) initializing model parameters in half precision instead of fp32 ([#21586](https://github.com/Lightning-AI/pytorch-lightning/pull/21586))
-
 - Fixed `device_mesh` type hint in `FSDPStrategy` to accept a 2-element tuple via the CLI ([#21581](https://github.com/Lightning-AI/pytorch-lightning/pull/21581))
 
 ---

src/lightning/pytorch/CHANGELOG.md

@@ -28,30 +28,26 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ---
 
-## [2.6.2] - 2026-03-19
+## [2.6.4] - 2026-05-20
 
+> Versions 2.6.2 and 2.6.3 were skipped due to a supply chain security compromise. See [#21691](https://github.com/Lightning-AI/pytorch-lightning/issues/21691) for details.
 
 ### Removed
 
 - Removed support for Neptune logger ([#21572](https://github.com/Lightning-AI/pytorch-lightning/pull/21572))
 
-
 ### Changed
 
 - Raise minimum `LitLogger` version to `2026-03-17` ([#21591](https://github.com/Lightning-AI/pytorch-lightning/pull/21591))
 
-
 ### Fixed
 
 - Fixed `val_check_interval` raising `ValueError` when `limit_val_batches=0` and interval exceeds training batches ([#21560](https://github.com/Lightning-AI/pytorch-lightning/pull/21560))
-
 - Fixed pkg-resources deprecation issue ([#21538](https://github.com/Lightning-AI/pytorch-lightning/pull/21538))
-
 - Fixed FSDP mixed precision (`bf16-mixed`, `16-mixed`) initializing model parameters in half precision instead of fp32 ([#21586](https://github.com/Lightning-AI/pytorch-lightning/pull/21586))
-
 - Fixed `device_mesh` type hint in `FSDPStrategy` to accept a 2-element tuple via the CLI ([#21581](https://github.com/Lightning-AI/pytorch-lightning/pull/21581))
-
-- Fixed ``RichModelSummary`` model size display formatting ([#21467](https://github.com/Lightning-AI/pytorch-lightning/pull/21467))
+- Fixed `RichModelSummary` model size display formatting ([#21467](https://github.com/Lightning-AI/pytorch-lightning/pull/21467))
+- Fixed floating-point precision in `SimpleProfiler` duration aggregation by using `math.fsum` ([#21525](https://github.com/Lightning-AI/pytorch-lightning/pull/21525))
 
 ---