Add deeper tree shapes to runtime microbenchmarks

[zach-klippenstein]

Summary

• Renamed ComposeRuntimeMicrobenchmark to WorkflowRuntimeMicrobenchmark to accurately reflect what's being benchmarked
• Added a SquareishTree shape (degree=3, depth=5, 243 leaves) alongside the existing ShallowBushyTree (degree=75, depth=1, 75 leaves) to better approximate real-world workflow trees
• Added extra work (side effects, remember calls) in intermediate and leaf workflows to more closely emulate real-world usage

Note: The two commits should be reviewed separately.

Here's an example run of the benchmarks on a Samsung phone (results parsed and formatted by Claude, but I spot-checked to numbers to confirm it's not hallucinating):

ShallowBushyTree (degree=75, depth=1, 75 leaves)

Benchmark	NoOptimizations	AllOptimizations
initialRenderAllChildren	1.49 ms	1.55 ms
initialRenderNewSibling	453.0 us	369.9 us
tearDownAllChildren	1.27 ms	1.38 ms
tearDownSingleSibling	464.7 us	379.8 us
rerenderAllChildrenByPropsChange	400.3 us	409.3 us
rerenderSingleSiblingByPropsChange	390.0 us	327.4 us
rerenderAllChildrenViaStateChange	25.34 ms	539.2 us (47x)
rerenderSingleSiblingViaStateChange	395.9 us	336.9 us

SquareishTree (degree=3, depth=5, 243 leaves)

Benchmark	NoOptimizations	AllOptimizations
initialRenderAllChildren	5.48 ms	5.83 ms
initialRenderNewSibling	1.67 ms	1.36 ms
tearDownAllChildren	5.50 ms	5.60 ms
tearDownSingleSibling	1.69 ms	1.38 ms
rerenderAllChildrenByPropsChange	1.54 ms	1.54 ms
rerenderSingleSiblingByPropsChange	1.56 ms	1.31 ms
rerenderAllChildrenViaStateChange	272.65 ms	4.05 ms (67x)
rerenderSingleSiblingViaStateChange	1.81 ms	1.05 ms

Test plan

• Ran all 32 benchmarks on a physical device (SM-A256U1, Android 16) — all passed
• Verified benchmark results are reasonable and consistent across runs

🤖 Generated with Claude Code