Bitops improvements

[brenns10]

In the spirit of (incrementally) reducing the verbosity of drgn, I noticed that the bitops APIs could use a bit of improvement.

>>> b = prog["boot_cpu_data"]
>>> # x86_capability is an array of __u32 ...
>>> bitfield = cast("unsigned long *", b.x86_capability[0].address_of_())
>>> bits = list(for_each_set_bit(bitfield, 8 * sizeof(b.x86_capability)))

First, the documentation indicated that they take an unsigned long *, so I assumed I needed to cast my array to a pointer, which is a pretty verbose operation. But reading the code, I could see that it worked perfectly fine on arrays, and also on pointers/arrays to any unsigned integer type. (And later, I noticed even the tests use an arrray rather than pointer). So this PR updates that documentation and explicitly tests a wider range of integer types (just to be sure). That way future users would know they can provide the bitfield directly without the casting.

Second, it seems like sized array types are relatively common in the kernel for bit fields (e.g. cpuinfo_x86.x86_capability link), and for these, we can infer the size parameter. So this PR makes it optional to provide that parameter.

Now with the clearer documentation and optional parameter, it's a lot easier:

>>> bits = list(for_each_set_bit(prog["boot_cpu_data"].x86_capability))

[osandov]

FWIW, this slightly less verbose version is equivalent to your cast:

cast("unsigned long *", prog["boot_cpu_data"].x86_capability)

But I like these improvements anyways, modulo my endianness concerns in the comments.