Replace rather than update Metal buffers

[TimSylvester]

With #1831, small non-index buffers (<= 4KiB) are encoded directly into the render command encoder rather than referenced as buffers. This should eliminate sequencing problems where the buffer is updated while the previous frame rendering is still using them.

In order to avoid the same problem with larger buffers (e.g., vertex attributes), this change makes buffer updates allocate a new Metal buffer instead of modifying an existing one. Index buffers were already being replaced rather than updating.

The effect on performance in the benchmarks seems minimal, but cases where the vertex buffers are large and changing regularly (e.g., symbols with layout in dense areas) there may be a more significant effect.

The previous buffers will remain allocated until they are no longer referenced, so memory utilization will increase somewhat.

Before:
Average frame encoding time: 2.6382 ms
Average frame rendering time: 14.5671 ms

Average frame encoding time: 2.6830 ms
Average frame rendering time: 14.5455 ms

Average frame encoding time: 2.5971 ms
Average frame rendering time: 14.5503 ms

---

After:
Average frame encoding time: 2.6585 ms
Average frame rendering time: 14.5019 ms

Average frame encoding time: 2.6152 ms
Average frame rendering time: 14.5189 ms

Average frame encoding time: 2.6410 ms
Average frame rendering time: 14.5200 ms

[sjg-wdw]

What a nice, small change.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (4150c8b) 85.83% compared to head (3cecfc7) 85.85%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1842      +/-   ##
==========================================
+ Coverage   85.83%   85.85%   +0.02%     
==========================================
  Files         568      568              
  Lines       27975    27986      +11     
==========================================
+ Hits        24012    24028      +16     
+ Misses       3963     3958       -5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Bloaty Results 🐋

Compared to main

    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +0.0%    +384  -0.0%     -16    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-1842-compared-to-main.txt

Compared to d387090 (legacy)

    FILE SIZE        VM SIZE    
 --------------  -------------- 
   +19% +22.4Mi  +403% +24.0Mi    TOTAL

Full report: https://maplibre-native.s3.eu-central-1.amazonaws.com/bloaty-results/pr-1842-compared-to-legacy.txt

[stefankarschti]

After some panning & zooming the street name labels are drawn in the wrong place. Maybe some UBO buffer is not getting updated and the matrix is wrong?

RPReplay_1842.MP4

[TimSylvester]

About 1/3 of buffers above the 4k threshold in the benchmark, but only 11K of 1.8M = 0.6% of updates are to large buffers.

totalBuffers = 48322
totalBufferObjs = 15239
bufferUpdates = 1769897
bufferObjUpdates = 10878
bufferUpdateBytes = 296601088
uniformUpdateBytes = 107493168