Simplify the protocol code

[jadamcrain]

Item 5 of #59.

Changes

• Object headers. One ObjectHeader.parse, matched on the qualifier, replaces ObjectHeaderParser and its seven parser objects. A start/stop header now parses its objects by count, stop − start + 1, so Variation.parse(range) is gone. A variation that takes neither header gives a single message for both: "may not be used with a start/stop or count header". The "may not by used" typo is also fixed in the two prefix messages.
• Little-endian decoding. Integers.fromBytesLE is now the only little-endian decoder. The field parsers, UInt16LE and the device attribute values all use it, and the byte-taking constructors of SInt16LE, SInt32LE, UInt32LE and UInt48LE are gone. One fixed-width field parser replaces the seven separate ones for the 16-, 32- and 48-bit integers and the floats.
• Point builders. There's now one builder per kind of value: boolean, double-bit, integer and floating-point. Each variation passes in its point type: an event variation's points are events, and every other variation's are static. That replaces the eight Static/Event factory pairs. Before the change I checked that every variation's old factory matched its object type. The typed copy_with_timestamp overrides are unchanged.
• Bit fields. BitFields packs and unpacks both single and double bits. It replaces BitFieldParser, DoubleBitFieldParser, DoubleBits, BitFieldBuilder and the now-unused Bits.
• Frames. TestDriver.toFrames is the one place a message becomes link frames. frameBytes, writeMessage and sendDatagrams all use it, and the order of logging is unchanged.

Checks

A differential test ran a seeded corpus of about 775,000 inputs before and after the change. The corpus has:

• every variation combined with every qualifier, an unknown qualifier, counts 0 to 3 and 0 to 30 random payload bytes
• events with relative times that follow a common time of occurrence

For each input it recorded the parsed headers, their contents and points, the re-serialized bytes, and the result of parsing the input as a response. Before and after are identical except for the error message above.