omindex: delay libmagic checks(ticket#743)

[caiyulun]

Link to the Ticket: https://trac.xapian.org/ticket/743

the libmagic call to get mime type is expensive, so we should check it later, for example, we can check the size by call the stat, check the timestamps and the DB for an existing entry before call libmagic.

This PR now achieves to check the filesize before call the libmagic.

To solved next:

since check the timestamps of a file need to check through the database in some situation: when the timestamps of the file return from stat call (says T1 for convenience) is older than the last time we run omindex, we need to look through the database to determine if the file is indexed and get the modify time when it is indexed (says T2 for convenience), if T2 is older than T1, we should re-index the file.

it is expensive to iterate through the DB, so we should figure out the order between libmagic check and timestamps in next step.

[ojwb]

Merged to git master as d32e135, thanks.

[ojwb]

One small but quite important point - you should aim to make comments clear and unambiguous, and getting the grammar and punctuation helps that (though I realise that's often harder for non-native speakers).

So better to capitalise the sentence, give the verb a subject ("we can't" rather than "can't"), put a space after the comma, and a full stop (period) at the end. I've pushed a commit to fix these in this case.