convert File History cache serialization to Smile

[vladak]

This change converts the serialization format of the file history cache to Smile (already used for annotation cache). Like always, this is a trade-off: for bigger cache there is a reduction of indexing speed. Also, this paves a way towards reducing the memory footprint of the webapp by fetching history in chunks, rather than whole (#3541). Also, it makes certain features like #4087 posible.

For benchmark, I used the Linux Git repository. Indexer options used for the test:

-P
-S
-H
-G
--progress
-s
/var/opengrok/src.linux
-d
/var/opengrok/data.linux
-W
/var/opengrok/etc/configuration.xml

_	size	time
before	696MB	0:42:18
after	1,9 GB	0:24:35

Tested on my trusty laptop with Intel Core i7 and built-in SSD.

Sizes reported by du -sh. So for almost 3x space the history cache creation speed is reduced by ~40 %. The space increase may sound like a lot, however one has to realize that not only the scheme was switched, but also there is no compression now. The lack of compression together with the switch allows for the stream based access. Given that some modern file systems support compression (e.g. per dataset in ZFS), it should not be a big deal. After uncompressing all the .gz files in the history cache, it went from 696 MB to 2,6 GB so obviously uncompressed XML has bigger footprint than Smile.

As for the implementation, the tags are stored in separate files, with the .t suffix. This makes it possible to desrialize the HistoryEntry objects as stream.

This also fixes some unrelated problems with progress reporting found when testing.