*   Added top-k to precision and recall to keras metrics.
    *   Add a ragged size op and register it to the op dispatcher
    *   Transitive dependencies on :`pooling_ops` were removed. Some users may
        need to add explicit dependencies on :`pooling_ops` if they reference
        the operators from that library.
    *   Add `CompositeTensor` base class.

        "//tensorflow/lite/toco:toco_flags_proto_cc",
        "//tensorflow/lite/toco:toco_graphviz_dump_options",
        "//tensorflow/lite/toco:toco_port",
        "//tensorflow/lite/toco:toco_tooling",
        "//tensorflow/lite/toco:tooling_util",
        "//tensorflow/lite/toco:types_proto_cc",
        "//tensorflow/lite:framework",
        "//tensorflow/lite:shared_library",
        "//tensorflow/lite:stateful_error_reporter",

h3. Plugins

* Execution model
* Metadata model
* Metadata tooling
 ** Metadata extractor
 ** Metadata reader
 ** Metadata writer
 ** Metadata adapter (if required for the target system)
* Maven packaging and lifecycle
* Maven test harness for plugin execution model
* Eclipse IDE tooling for plugin execution model and metadata model

- we also seem to have information like the plugin lifecycle model that's buried inside the maven execution model

- Failed pods associated with a job with `parallelism = 1` are recreated by the job controller honoring exponential backoff delay again. However, for jobs with `parallelism > 1`, pods might be created without exponential backoff delay. ([#115020](https://github.com/kubernetes/kubernetes/pull/115020), [@nikhita](https://github.com/nikhita)) [SIG Apps]...

  - `setcap` to v2.0.3 ([#103235](https://github.com/kubernetes/kubernetes/pull/103235), [@thejoycekung](https://github.com/thejoycekung)) [SIG API Machinery, Release and Testing]

### Bug or Regression

- Failed pods associated with a job with `parallelism = 1` are recreated by the job controller honoring exponential backoff delay again. However, for jobs with `parallelism > 1`, pods might be created without exponential backoff delay. ([#115021](https://github.com/kubernetes/kubernetes/pull/115021), [@nikhita](https://github.com/nikhita)) [SIG Apps]...

  - `setcap` to v2.0.3 ([#103235](https://github.com/kubernetes/kubernetes/pull/103235), [@thejoycekung](https://github.com/thejoycekung)) [SIG API Machinery, Release and Testing]

### Bug or Regression

- Fix kube-proxy regression on UDP services because the logic to detect stale connections was not considering if the endpoint was ready. (#106239, @aojea) [SIG Network and Testing]
- Fix scoring for NodeResourcesBalancedAllocation plugins when nodes have containers with no requests. (#106081, @ahmad-diaa) [SIG Scheduling]
- Support more than 100 disk mounts on Windows (#105673, @andyzhangx) [SIG Storage and Windows]

- Failed pods associated with a job with `parallelism = 1` are recreated by the job controller honoring exponential backoff delay again. However, for jobs with `parallelism > 1`, pods might be created without exponential backoff delay. ([#115022](https://github.com/kubernetes/kubernetes/pull/115022), [@nikhita](https://github.com/nikhita)) [SIG Apps]...

 * b0
 * --
 *
 * The inclusive start of the range. We get the first 14 bits of this code point from the section
 * and the last 7 bits from this byte.
 *
 * The end of the range is not encoded, but can be inferred by looking at the start of the range
 * that follows.
 *
 * b1
 * --
 *
 * This is either a mapping decision or the length of the mapped output, according to this table:
 *
 * ```