Add profiling setup using geodepot

[balazsdukai]

Overview

This branch splits into two main themes:

1. profiling infrastructure for repeatable large-input runs
2. pre-write performance work in the CityJSON-to-GLB pipeline

It also adds supporting documentation for the profiling workflow, the performance investigation, and the associated optimization decision record.

Profiling Workflow

• Added a repo-local just profile entrypoint and scripts/profile-tyler.sh.
• Integrated profiling by Geodepot case name, where the case spec resolves both:
  • the input dataset path via geodepot get <case spec>
  • the case-specific profile-tyler.json data item via the case root
• Added support for selecting which profiler runs:
  • --runner perf
  • --runner massif
  • --runner all by default
• Preserved failed profiling runs so long Massif sessions can be inspected without rerunning them.
• Wrote structured per-run output under docs/performance/runs/ with:
  • manifest.json
  • perf-stat.json
  • massif-summary.json
  • summary.md

Performance Changes

• Reduced allocations in the grid-counting and tile-preparation path.
• Added a faster path for direct unique cell assignment in the grid counter.
• Skipped no-op tile-preparation work when it cannot affect GLB output.
• Added timing and count logging around the main hot spots so future profiling is easier to interpret.
• Improved glTF writer and CLI logging to support debugging and profiling.