[0.9.3] invalid timestamp format should throw parser error, not be ignored

[beckettsean]

When giving explicit timestamps in queries they must be single-quoted:

> select count(value) from cpu_idle where time >= '2015-08-25T20:00:00Z' and time <= '2015-08-25T21:00:00Z'  group by time(10m)
name: cpu_idle
--------------
time            count
2015-08-25T20:00:00Z    8
2015-08-25T20:10:00Z    144
2015-08-25T20:20:00Z    688
2015-08-25T20:30:00Z    688
2015-08-25T20:40:00Z    688
2015-08-25T20:50:00Z    680
2015-08-25T21:00:00Z    688

Double-quoting returns results as if there were no time restriction at all:

> select count(value) from cpu_idle where time > "2015-08-25T20:00:00Z" and time < "2015-08-25T21:00:00Z"  group by time(10m)
name: cpu_idle
--------------
time            count
1970-01-01T00:00:00Z    618464

> select count(value) from cpu_idle
name: cpu_idle
--------------
time            count
1970-01-01T00:00:00Z    618464

[corylanou]

We should specifically return an error if we detected a literal (double quotes instead of single) and error with something like time literals need to be single quoted, not double quoted.

[beckettsean]

Per Paul, in a WHERE time clause, first try to parse the double-quoted value as a timestamp, not an identifier. If that fails fall back to parsing it as an identifier.

[jsternberg]

The reason for this problem is because the function that retrieves the time range doesn't send back an error when it finds an objectionable time comparison like was done here.

The function that needs to be changed is influxql.TimeRange(). I can probably do this one with #3290.