Moving code to extract data from byte slices into bytesconv.go.

Setting tx levels within a group is not implemented yet. The code
lacks error checking, and is presently rather dreadful. But tests
pass.

strconv.Atoi(string([]byte)) causes the temporary string to escape
to the heap. This way we stay with a byte slice that does not escape.

VSLQ_Dispatch() is a powerful tool for the C API, but the restrictions
and overhead involved with the C/Go bridge make it too unwieldy to use.
Too much synchronization becomes necessary for a low-level API like
this. The cgocall overhead becomes excessive.

It seems to be difficult, perhaps impossible, to implement callbacks
without creating many objects on the Go heap. Since many objects from
the log are created in hot code, this puts excessive pressure on the
Go garbage collector.

This version uses low-level log read operations, excusively VSL_Next()
and related functions. The C/Go bridge is much smaller and simpler. It
should be possible to create the objects for log data on the Go stack.

The Cursor and Query interfaces reflect the underlying VSL interface
more faithfully. Distinct Cursors are safe for concurrent reads (as
for struct VSL_cursor from the VSL interface).

This means that grouping and VSL queries will have to re-implemented
in Go.