i think i might have confused myself when implementing --failed-list and ultimately tried to have it do double duty:
(1.1) when doing an initial upload of symbols, --failed-list points to a write-only file where the uploading process writes out paths that failed to upload
(1.2) when trying to resume a failed upload, --failed-list points to a read-only file that the uploader should use to upload a reduced set of files

the idea being that the flow is:
(2.1) bot runs upload_symbols after a new build and saves all upload failures to the "failed list"
(2.2) bot uploads the "failed list" and the full sym archive
(2.3) later on, a dev runs upload_symbols by hand using the "failed list" and the full sym archive to only upload the symbol files that failed the first time around

it looks like i only ever implemented the (1.1) feature, and then got confused and thought (1.2) just worked.  but that isn't the case, so when trying to do (2.3) above, we actually upload the entire archive again and ignore the failed list as an input filter.

i think what we want to do is add a new flag (maybe "--resume-list", feel free to bikeshed) that we'd use in (2.3) so that the code clearly can differentiate between "filter input file" and "output log file".