Yes, there's a good argument that the license would apply automatically to modifications to an existing file, but even in that case it is somewhat questionable. Is the modification substantial enough that it could be considered a new, stand-alone, derived work? If so, the pre-existing license may not apply. Was the modification submitted as a patch in a file format that does not incorporate all of the pre-existing code? If so, it may be a new, stand-alone work to which the pre-existing license does not apply. Etc., etc. Very little of this stuff has been well tested in court, so there's a lot of room for facially plausible legal challenges.
And it's fairly clear that new files are new works. So, really, in all cases, unless you're getting an explicit statement from contributors that they have placed their contributions under the relevant license, there will be a a lot of uncertainty about whether they really have been—particularly if the contributor (or his or her employer) later challenges your claim.
That's why Github and other platforms for projects on the pull-request model should really build explicit consent to licensing into the process.
And it's fairly clear that new files are new works. So, really, in all cases, unless you're getting an explicit statement from contributors that they have placed their contributions under the relevant license, there will be a a lot of uncertainty about whether they really have been—particularly if the contributor (or his or her employer) later challenges your claim.
That's why Github and other platforms for projects on the pull-request model should really build explicit consent to licensing into the process.