This was an option in the original proposal, but was (vigorously)
debated and ultimately removed, because
* the boundary is a reliable indicator of the extent of the file data
* if both content-length and boundary are present, you have to check
the boundary anyway.
* in some situations computing the content-length is difficult
(although necessary for HTTP POST in some situations).
That is, while developers might think it would be easier with explicit
content-length parts, they're wrong.
The RFC should probably have said this explicitly.