Some tcpinfo rows too large.

[gfr10598]

Inserts are sometimes failing on tcpinfo buffers. Likely due to large number of snapshots for some rows.

Should make two changes:

1. Limit number of snapshots in a row. Perhaps filter down to 10/second after the first 12 seconds, and 1/second after the first 60 seconds?
2. Handle insert errors by submitting rows individually.

[gfr10598]

#2 is done. The buffer is now split in half and retried. Only single row failures are not retried.