Replace HistoryQuickProvider protobuf-based caching with an SQLite-based database.

Replace HistoryQuickProvider protobuf-based caching with an SQLite-based database.

The protobuf-based cache was only being read at startup and written at shutdown. (Except that when the cache could not be read at startup the cache would immediately be saved upon private data reconstruction from the History database.) With the new implementation using an SQLite database for the HQP cache the private data will be restored from the cache database at startup and continually updated during normal operation. There is no need to flush or write the cache at shutdown as it is constantly kept up-to-date.

Detailed comments about the changes made in this CL can be found in the document shared separately.

Previous reviewers: sky, pkasting, shess.

BUG=95686, 95876, 131668
TEST=New tests added, old tests updated, all tests pass.
To manually verify changes:
NOTE: For tests using chrome://omnibox be sure to check the "Show results per provider" and then look at the results for the HistoryQuickProvider.
1) New visits: Type an URL never visited before. Bring up new tab. Start typing the URL or parts of the page title or both and verify that the page is offered as a suggestion.
2) New visits: Type an URL never visited before. Do search using chrome://omnibox. New visit should show.
3) Delete visit: Visit some pages and verify they have been recorded. Bring up history and delete one of the visits. Check via chrome://omnibox that it was deleted.
4) Clear history: Visit some pages. Clear the visit history. Check via chrome://omnibox that none of the visits are still returned by the HQP.
5) Updated site: Create a local page and visit it. Search using chrome://omnibox by page title and verify it can be found. Change the page title dramatically and revisit the page. Verify via chrome://oomnibox that the page can be found by words from the new title.

Previously reviewed as: http://codereview.chromium.org/10477018/.
10477018 was reverted due to memory leaks detected by build bots.
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=152946

[mrossetti, 2012-08-22 19:04:18]

You may have heard that there were a dozen or more leaks reported by the
linux_valgrind and linux_chromeos_valgrind bots. I decided to pull my changes
and try to eliminate as many leaks as possible. They were all related to test
shutdown and I suppose I could have just added a bunch of suppressions but I was
uncomfortable with that.

You have already reviewed the bulk of the changes shown in this CL. I would
suggest only looking at those files that have changes against patch set #1. You
will see that the only substantial changes are in
InMemoryURLIndexBaseTest::TearDown():

1) Letting the outstanding history operations flush in unit test TearDown().
2) Using a WaitableEvent to insure the DB thread has been flushed in unit test
TearDown().
3) Reordering a bit the thread waitage in unit test TearDown().
4) Adding three suppressions for the three remaining leaks. I'd like to handle
these leaks in separate CLs as they are clearly related only to unit tests. I
believe what I need to do is to rearrange the thread shutdown flushing
coordination on a unit test basis rather than solely at the unit test harness
level (i.e. only in InMemoryURLIndexBaseTest::TearDown()).

There was one other significant change in 

5) Adjust the shutdown lockage a bit to better guard the shutdown_ member. (I'm
going to look at using a CancellationFlag for handling Shutdown() handling in a
future CL.)

If you're okay with this then I'm going to commit.

[sky, 2012-08-22 19:42:34]

LGTM

http://codereview.chromium.org/10837244/diff/36006/chrome/browser/history/url...
File chrome/browser/history/url_index_private_data.cc (right):

http://codereview.chromium.org/10837244/diff/36006/chrome/browser/history/url...
chrome/browser/history/url_index_private_data.cc:655: return shutdown_;
nit: spacing.