`length` / `bit_length` / `octet_length` fall back to Spark on binary input

Part of #5572.

`length`, `bit_length` and `octet_length` all decline `BinaryType` input, and each falls the whole projection back to Spark:

```scala
case _: BinaryType => Unsupported(Some("Length on BinaryType is not supported"))
```

`spark/src/main/scala/org/apache/comet/serde/strings.scala:89` (`CometLength`), `:98` (`CometBitLength`), `:107` (`CometOctetLength`)

All three are trivial in Spark's generated code — `length(binary)` is `.numBytes()`, and the other two are arithmetic on it. `BinaryType` is in `CometBatchKernelCodegen.isSupportedDataType`, so the dispatcher handles them.

Fix: mix `CodegenDispatchFallback` into all three. They are near-identical serdes, so this is one small change.

The generated docs list all three as "✅ Native" with no note today, so they should pick up the Hybrid marker once the mixin is added.