Unclear requirements for CL_DEVICE_DOUBLE_FP_CONFIG

The table of device queries says
```
If double precision is supported by the device,
then the minimum double precision floating-
point capability must be:
CL_FP_FMA |
CL_FP_ROUND_TO_NEAREST |
CL_FP_INF_NAN |
CL_FP_DENORM.
```

but the text for the `-cl-denorms-are-zero` build options says:

```
This option is ignored for double precision numbers if the device does not support double
precision or if it does support double precision but not double precision denormalized numbers
i.e. CL_FP_DENORM bit is not set in CL_DEVICE_DOUBLE_FP_CONFIG.
```
The former requires denormals to always be supported, the latter suggest that it's not always the case.

Start by reviewing the device-query table and the `-cl-denorms-are-zero` build-option text quoted in the issue. Determine which requirement for `CL_DEVICE_DOUBLE_FP_CONFIG` is authoritative and update the OpenCL documentation so the two statements agree. Done means the denormal support requirement is stated consistently.

Written by the indexing model from the issue text.