Git push can indefinitely block "waiting for journal bytes to flush"

On one occasion:

$ git push kb master
Initializing Keybase... done.
Syncing with Keybase... done.
Counting and decrypting: 388 objects... done.
Preparing and encrypting: (100.00%) 388/388 objects... done.
Indexing hashes: (100.00%) 388/388 objects... done.
Indexing CRCs: (100.00%) 388/388 objects... done.
Indexing offsets: (100.00%) 388/388 objects... done.
Syncing encrypted data to Keybase: (0.00%) 0.00/20.89 KB...

...waited indefinitely at that "0.00%" twice, shortly after the repository in question, succeeding on later 3rd try.

Logs from the second failed instance look like:

2017-10-08T15:45:15.432523-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:1576] 3006 SetInitialHeadFromServer, revision=15 (unmerged) [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.432612-05:00 ▶ [DEBU git(PRE) prefetcher.go:218] 3007 Prefetching block by request from upstream component. Priority: 1073741824
2017-10-08T15:45:15.432683-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:1602] 3008 Head MD already set to revision 15 (unmerged), no need to set initial head again [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.432732-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:1604] 3009 SetInitialHeadFromServer, revision=15 (unmerged) done: <nil> [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.432777-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:1717] 300a getRootNode [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.432845-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:1750] 300c getRootNode done: NodeID(0xc42020c800) <nil> [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.432935-05:00 ▶ [DEBU git kbfs_ops.go:510] 300d Done: <nil> [tags:GITCMDID=7LXHNfvEcBbcSoPmRO5Skg,GITID=vF6E-T-EssErOZNECCPcsQ]
2017-10-08T15:45:15.433014-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:4201] 300e SyncAll [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.432802-05:00 ▶ [WARN git block_retrieval_queue.go:314] 300b Block missing for disk block cache metadata update
2017-10-08T15:45:15.433148-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:2769] 300f Blocking until squash finishes [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.433199-05:00 ▶ [DEBU git(FBO bf3b6829) folder_branch_ops.go:4208] 3010 SyncAll done: <nil> [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]
2017-10-08T15:45:15.433258-05:00 ▶ [DEBU git runner.go:440] 3011 Waiting for 21391 journal bytes to flush [tags:GITID=vF6E-T-EssErOZNECCPcsQ,GITCMDID=7LXHNfvEcBbcSoPmRO5Skg]

Start by tracing the journal flush wait in runner.go:440 and the preceding SyncAll path in folder_branch_ops.go, then compare it with the block-retrieval warning in block_retrieval_queue.go:314. Use the supplied logs to determine why the journal does not finish flushing and reproduce the intermittent Git push hang. Done means the push no longer blocks indefinitely and a regression test covers the failure.