Proteins are essential components for molecular machines, so they represent predefined general targets in amino acid sequence space. A minimum set required to create a molecular machine or other structure would be a more specific target. These targets might never be realized, or they might be found in some organism at some time in history. The challenge is that finding even one protein of a size comparable to a larger flagellar protein would require for 1E132 distinct protein possibilities to exist. I would be interested to see any expert who has ever claimed they are that abundant.
In addition, for evolution to serendipitously stumble across complex molecular machines, the search space would have to be filled with functional possibilities. In other words, vast numbers of other possible configurations of proteins corresponding to locomotive machines must be possible for any to be found. However, if so many possibilities exist, the chances of finding a solution which matches a human-designed outboard rotary motor so closely is nil.
However, if so few solutions exist in both the search space for human engineering and for biological systems that such similar designs appear in both engineering and biology, then the chances of nature stumbling upon one of them is nil, since functional combinations would be so rare. In summary, the fact that the flagellum matches so closely with a known design logic means that it can better be understood as a predefined target corresponding to a rotary motor than a haphazardly fit together collection of proteins which just happened to function as a motor.
Natural selection cannot help add even a single piece to the motor, such as the filament, since adding the piece requires multiple highly-specialized proteins. Until all appear and are then fully integrated into the assembly process, natural selection would work against adding the filament. The production of a useless protein wastes resources and would be quickly disabled through mutations.
You might try to argue that each protein (filament, assembly cap, 2 joint proteins) served some other role, but the degree of specialization makes such a possibility difficult to justify. And, even if they did play some specific role in some other context, selection would prevent them from being redirected toward a completely different process.
A useful exercise would be attempting to devise or find a semi-detailed scenario for the evolutionary development of any complex, highly-integrated multi-protein structure. Any realistic attempt would run into these challenges.