[sql] Use memory-mapped I/O for sql::Connection.

[sql] Use memory-mapped I/O for sql::Connection.

sql::Connection::Open*() uses PRAGMA mmap_size to enable SQLite's
memory-mapped I/O.  Additionally instrument to flush dirty pages from
the page cache after writes.

BUG=489784, 533682, 537681, 537636
[Relands https://crrev.com/9a1948a4d6d4 (#350362),
reverted at https://crrev.com/55c3216b15ce (#350386),
relanded at https://crrev.com/5c2f4e50d96d (#351344),
reverted at https://crrev.com/745e18c52384 (#351596).
May the waterfall have mercy on my CL.]

Committed: https://crrev.com/7dbd4dee4702e30efb24c81898ae45dd0d502c41
Cr-Commit-Position: refs/heads/master@{#352631}

[Scott Hess - ex-Googler, 2015-10-02 23:23:52]

Disable mmap test under mojo.

[Scott Hess - ex-Googler, 2015-10-02 23:24:18]

Fix uninit vars in Connection.

[Scott Hess - ex-Googler, 2015-10-02 23:24:47]

Check for open db before releasing cache.

[Scott Hess - ex-Googler, 2015-10-02 23:36:54]

https://codereview.chromium.org/1349863003/, slipstream
https://codereview.chromium.org/1355403004/ to fix mojo test, slipstream
https://codereview.chromium.org/1368493002/ to fix valgrind complaint,
slipstream
https://codereview.chromium.org/1381563003/ to fix http://crbug.com/537636

[Scott Hess - ex-Googler, 2015-10-02 23:38:15]

On 2015/10/02 23:36:54, Scott Hess wrote:
> https://codereview.chromium.org/1349863003/, slipstream
> https://codereview.chromium.org/1355403004/ to fix mojo test, slipstream
> https://codereview.chromium.org/1368493002/ to fix valgrind complaint,
> slipstream
> https://codereview.chromium.org/1381563003/ to fix http://crbug.com/537636

https://codereview.chromium.org/1368293012 landed separately to fix tsan issue
in http://crbug.com/537681

[Scott Hess - ex-Googler, 2015-10-02 23:42:43]

shess@chromium.org changed reviewers:
+ pavely@chromium.org, rmcilroy@chromium.org

[Scott Hess - ex-Googler, 2015-10-02 23:42:44]

Back for more!

pavely@ no changes to what you previously LGTM'ed in sync.
rmcilroy@ ps4 fixes the crash seen the night it was on production, the cause of
the revert was fixed under another CL.

About one more bounce and I'll land the skeleton of things so that I can start
bouncing a one-line CL.  At this point I'm terrified that the next bounce will
be because of a stupid merge mistake.

https://codereview.chromium.org/1382283003/diff/50001/sql/connection.cc
File sql/connection.cc (right):

https://codereview.chromium.org/1382283003/diff/50001/sql/connection.cc#newco...
sql/connection.cc:1370: // some cases sql::Transaction can fire rollback after a
database is closed.
This test _could_ be at the top of this method, but the ROLLBACK itself could
cause the error handler to close things.

[pavely, 2015-10-02 23:48:30]

sync/* lgtm

[rmcilroy, 2015-10-06 10:29:19]

lgtm.

[Scott Hess - ex-Googler, 2015-10-06 16:24:25]

On 2015/10/06 10:29:19, rmcilroy wrote:
> lgtm.

Thanks.  I think the trybot failures were all unrelated to previous failures, so
....

[Scott Hess - ex-Googler, 2015-10-06 16:24:29]

The CQ bit was checked by shess@chromium.org

[commit-bot: I haz the power, 2015-10-06 16:25:05]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1382283003/50001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1382283003/50001

[commit-bot: I haz the power, 2015-10-06 17:39:25]

Committed patchset #4 (id:50001)

[commit-bot: I haz the power, 2015-10-06 17:40:15]

Patchset 4 (id:??) landed as
https://crrev.com/7dbd4dee4702e30efb24c81898ae45dd0d502c41
Cr-Commit-Position: refs/heads/master@{#352631}