*   Added a distributed runtime using GRPC
*   Move skflow to `contrib/learn`
*   Better linear optimizer in `contrib/linear_optimizer`
*   Random forest implementation in `contrib/tensor_forest`
*   CTC loss and decoders in `contrib/ctc`
*   Basic support for `half` data type
*   Better support for loading user ops (see examples in `contrib/`)
*   Allow use of (non-blocking) Eigen threadpool with

01BC          ; mapped                 ; 01BD          # 1.1  LATIN CAPITAL LETTER TONE FIVE
01BD..01C3    ; valid                                  # 1.1  LATIN SMALL LETTER TONE FIVE..LATIN LETTER RETROFLEX CLICK
01C4..01C6    ; mapped                 ; 0064 017E     # 1.1  LATIN CAPITAL LETTER DZ WITH CARON..LATIN SMALL LETTER DZ WITH CARON

`indices[n][j]` for `(m,i) < (n,j)` the slice `data[n][j]` will appear in the
merged result. If you do not need this guarantee, ParallelDynamicStitch might
perform better on some devices.

For example:

```python
    indices[0] = 6
    indices[1] = [4, 1]
    indices[2] = [[5, 2], [0, 3]]
    data[0] = [61, 62]
    data[1] = [[41, 42], [11, 12]]

		},
		numErrs: 0,
	}, {
		// For now we open firewalls, and its insecure if we open 10250, remove this
		// when we have better protections in place.
		name: "invalid port type=LoadBalancer",
		tweakSvc: func(s *core.Service) {
			s.Spec.Type = core.ServiceTypeLoadBalancer

		{"Flag.Usage", Field, 0},
		{"Flag.Value", Field, 0},
		{"FlagSet", Type, 0},
		{"FlagSet.Usage", Field, 0},
		{"Float64", Func, 0},
		{"Float64Var", Func, 0},
		{"Func", Func, 16},
		{"Getter", Type, 2},
		{"Int", Func, 0},
		{"Int64", Func, 0},
		{"Int64Var", Func, 0},
		{"IntVar", Func, 0},
		{"Lookup", Func, 0},
		{"NArg", Func, 0},
		{"NFlag", Func, 0},
		{"NewFlagSet", Func, 0},

http://physics.nist.gov/cuu/Units/binary.html)\n\n<decimalSI>       ::= m | \"\" | k | M | G | T | P | E\n\n\t(Note that 1024 = 1Ki but 1000 = 1k; I didn't choose the capitalization.)\n\n<decimalExponent> ::= \"e\" <signedNumber> | \"E\" <signedNumber> ```\n\nNo matter which of the three exponent forms is used, no quantity may represent a number greater than 2^63-1 in magnitude, nor may it have more than 3 decimal places. Numbers larger or more precise will be capped or rounded up. (E.g.: 0.1m will rounded up to...

http://physics.nist.gov/cuu/Units/binary.html)\n\n<decimalSI>       ::= m | \"\" | k | M | G | T | P | E\n\n\t(Note that 1024 = 1Ki but 1000 = 1k; I didn't choose the capitalization.)\n\n<decimalExponent> ::= \"e\" <signedNumber> | \"E\" <signedNumber> ```\n\nNo matter which of the three exponent forms is used, no quantity may represent a number greater than 2^63-1 in magnitude, nor may it have more than 3 decimal places. Numbers larger or more precise will be capped or rounded up. (E.g.: 0.1m will rounded up to...