*   Added gradient for `SparseToDense` op.
    *   Expose a flag that allows the number of threads to vary across Python
        benchmarks.
    *   `image.resize` in 2.0 now supports gradients for the new resize kernels.
    *   `image.resize` now considers proper pixel centers and has new kernels
        (incl. anti-aliasing).
    *   Renamed `tf.image` functions to remove duplicate "image" where it is
        redundant.

# MACOS ARM64 WHEEL

                val parent = fullFqName.parent()
                return resolveKdocFqName(analysisSession, selectedFqName, parent, contextElement)
            }
            val goBackSteps = fullFqName.pathSegments().size - selectedFqName.pathSegments().size
            check(goBackSteps > 0) {
                "Selected FqName ($selectedFqName) should be smaller than the whole FqName ($fullFqName)"
            }

Instead, an Istio specific field like `ExpectedTLSIdentity: spiffe://foo.bar` can encode the same information, at a fraction of the cost.

In our testing, even the most generous representations give custom types a 10x edge (in size, allocations, and CPU time) over Envoy types.
In addition, they are more clear and strictly typed; using Envoy types would require us to put a lot of information in untyped `metadata` maps.

   */
  internal fun nextConnectionSpec(
    connectionSpecs: List<ConnectionSpec>,
    sslSocket: SSLSocket,
  ): ConnectPlan? {
    for (i in connectionSpecIndex + 1 until connectionSpecs.size) {
      if (connectionSpecs[i].isCompatible(sslSocket)) {
        return copy(connectionSpecIndex = i, isTlsFallback = (connectionSpecIndex != -1))
      }
    }
    return null
  }

- In the "File already exists" dialogue, choose "Yes" to overwrite
- In the "Open Project" dialogue, choose "Delete Existing Project and Import"
- Revert the Git changes to files in the `.idea` folder

NOTE: Due to the project size, the very first import can take a while and IntelliJ might become unresponsive for several seconds during this period.

    val settings1 = Settings()
    settings1[Settings.HEADER_TABLE_SIZE] = 10000
    settings1[Settings.INITIAL_WINDOW_SIZE] = 20000
    settings1[Settings.MAX_FRAME_SIZE] = 30000
    peer.sendFrame().settings(settings1)
    peer.acceptFrame() // ACK SETTINGS
    val settings2 = Settings()
    settings2[Settings.INITIAL_WINDOW_SIZE] = 40000
    settings2[Settings.MAX_FRAME_SIZE] = 50000
    settings2[Settings.MAX_CONCURRENT_STREAMS] = 60000

| `resident_memory_bytes`       | `gauge`   | Resident memory size in bytes                                                                                  | `server` |
| `virtual_memory_bytes`        | `gauge`   | Virtual memory size in bytes                                                                                   | `server` |

| `minio_node_process_resident_memory_bytes` | Resident memory size in bytes.                                                                                  |
| `minio_node_process_virtual_memory_bytes`  | Virtual memory size in bytes.                                                                                   |

LINUX_AGENT_BINARIES:=./cni/cmd/istio-cni \
  ./cni/cmd/install-cni \
  $(AGENT_BINARIES)

BINARIES:=$(STANDARD_BINARIES) $(AGENT_BINARIES) $(LINUX_AGENT_BINARIES)

# List of binaries that have their size tested
RELEASE_SIZE_TEST_BINARIES:=pilot-discovery pilot-agent istioctl envoy ztunnel client server

# agent: enables agent-specific files. Usually this is used to trim dependencies where they would be hard to trim through standard refactoring