Batch up scheduling of deferred deletions

[TimSylvester]

In each tile cache (one per tile source), released tiles are kept in a vector and scheduled for release all at once, at the end of update, which is called once per frame.

The capture below is from ~150 seconds of the Android benchmark, 5300 frames. Of 384 total, it shows 138 instances of just one tile being scheduled, up to 3 instances of 13 being released at once. 440 total scheduled tasks were averted in this case, roughly half.

Manually zooming in and out on the same area for ~600 frames produces similar results.

If we only did the release after accumulating several tiles, we could reduce the number of tasks significantly more, but we might end up holding onto unused tiles for many frames in some cases.

Additionally, this avoids allocating a shared_ptr for each unique_ptr to make it copyable along with the lambda capture, using a more generic move-only function wrapper.

[github-actions]

Bloaty Results (iOS) 🐋

Compared to main

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.2% +24.5Ki  +0.1% +16.0Ki    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results-ios/pr-3030-compared-to-main.txt

[github-actions]

Bloaty Results 🐋

Compared to main

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.1% +79.0Ki  +0.0% +13.0Ki    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-3030-compared-to-main.txt

Compared to d387090 (legacy)

    FILE SIZE        VM SIZE    
 --------------  -------------- 
   +30% +34.6Mi  +433% +25.8Mi    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-3030-compared-to-legacy.txt

[github-actions]

Benchmark Results ⚡

Benchmark                                                     Time             CPU      Time Old      Time New       CPU Old       CPU New
------------------------------------------------------------------------------------------------------------------------------------------
OVERALL_GEOMEAN                                            +0.0013         +0.0010             0             0             0             0

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/benchmark-results/pr-3030-compared-to-main.txt

[louwers]

For future reference: once C++23 is out we can use std::move_only_function