[core] Mark used offline region resources in batches

[pozdnyakov]

Marking offline region resources "used" is on the hot code path for offline region download.
Previously, the engine attempted to update the database separetely for each resource and it caused a huge performance impact, especially when some tiles were already loaded.

Now, offline region resources are marked in batches (200 items at once).

Consider the time mesaurements of the mbgl-offline tool execution (Linux x64, release build):

time ./mbgl-offline --north 41.4664 --west 2.0407 --south 41.2724 --east 2.2680 --output barcelona.db

without patch:

1st time:

    real      0m9,505s
    user      0m3,315s
    sys       0m0,378s

2nd time (tiles are already loaded):

   real      0m9,480s
   user      0m0,403s
   sys       0m0,645s

with the patch:

1st time:

    real      0m7,399s
    user      0m2,994s
    sys       0m0,384s

2nd time (tiles are already loaded):

    real      0m0,149s
    user      0m0,059s
    sys       0m0,044s

So, the proposed changes make the repeated load ~64 times faster 🎉

[alexshalamov]

lgtm

[alexshalamov]

@LukasPaczos would you be able to test this PR for Android SDK?

[LukasPaczos]

The improvements are amazing!

A small offline region that takes about 8s to download:

	Previously	Now
First re-download in the current process	8s	1.3s
Subsequent re-downloads in the same process	4.5s	0.7s

What's interesting, first re-download in the app process (for example after restarting the application when tiles are already in place) takes significantly longer than subsequent ones.

[andrewharvey]

Wow I didn't realise you could call mbgl-offline twice on the same output and it would just update, I always thought it would just replace it and re-download everything. Could we update this in the documentation? PS. I can't find any of the offline sideload documentation anymore, I guess it must have been removed on purpose.

[LukasPaczos]

A region with 5001 resources that originally takes ~47s to download:

Testing different batch sizes https://github.com/mapbox/mapbox-gl-native/pull/15521/files#diff-72890ff2093a9bfa6bfe3a9e8955d6dfR361:

For batches of 200 resources (as in the PR)

	Previously	Now
First re-download in the current process	41s	2.7s
Subsequent re-downloads in the same process	38s	1.3s

For batches of 1000 resources

	Previously	Now
First re-download in the current process	41s	1.6s
Subsequent re-downloads in the same process	38s	1.0s

[pozdnyakov]

We could make batch size adaptive, i.e. depend on the total resources count, I'll investigate it and implement in a follow-up PR.

[chloekraw]

Wow, amazing work @pozdnyakov!

cc @zugaldia @tmpsantos