Add spdlog logging and UTL profiling to lidar odometry hot path

[bloom256]

Summary

• Add spdlog logging library to replace std::cout/std::cerr in lidar_odometry_utils_optimizers.cpp with structured, level-based logging (spdlog::info, spdlog::warn, spdlog::debug)
• Replace manual std::chrono timing with spdlog::stopwatch for cleaner timer code
• Add UTL profiler instrumentation to compute_step_2 and optimize_lidar_odometry hot paths for performance analysis, controlled by UTL_PROFILER_DISABLE CMake option (disabled by default for zero overhead)
• Add spdlog (v1.17.0) and UTL (v9.0.0) as git submodules under 3rdparty/

Note: Currently spdlog is only used in lidar_odometry_utils_optimizers.cpp. The remaining source files still use std::cout/`std::cerr. I suggest gradually migrating the rest of the project to spdlog in follow-up PRs.

Note: - Profiler macros are kept in the code - the benefit is you can enable profiling on any build just by flipping one CMake option, no code changes, no risk of forgetting to re-add instrumentation points

Details

spdlog integration

• Replaced all active std::cout/std::cerr statements in lidar_odometry_utils_optimizers.cpp with spdlog::info() and spdlog::warn()
• Replaced 4 manual std::chrono timing patterns with spdlog::stopwatch (stopwatch_total, stopwatch_worker, stopwatch_realtime, stopwatch_update)
• Initialized spdlog in main() with spdlog::cfg::load_env_levels() (runtime level control via SPDLOG_LEVEL env var) and spdlog::flush_on(spdlog::level::warn)
• Linked spdlog::spdlog to all targets that compile lidar_odometry source files (step1, pybind, 4 dependent apps)

Enabling debug logging at runtime:

Set the SPDLOG_LEVEL="debug" environment variable before running the application. No recompilation needed.

# Linux/macOS
SPDLOG_LEVEL=debug ./lidar_odometry_step_1 ...

# PowerShell (Windows)
$env:SPDLOG_LEVEL = "debug"
.\lidar_odometry_step_1.exe ...

Available levels (from most to least verbose): trace, debug, info, warn, err, critical, off. Default level is info.

UTL profiler instrumentation

Profiler segments added to the optimization hot path. The resulting profile looks like this:

-------------------- UTL PROFILING RESULTS ---------------------

# Thread [main] (reuse 0) (joined) (runtime -> 79459.01 ms)
   - 62.43% ------ | 49602.99 ms |          compute_step_2 | lidar_odometry_utils_optimizers.cpp:2245, compute_step_2()          |
     - 2.21% ----- |  1759.80 ms |       before_iterations | lidar_odometry_utils_optimizers.cpp:2310, compute_step_2()          |
     - 44.12% ---- | 35059.92 ms |          iteration_loop | lidar_odometry_utils_optimizers.cpp:2505, compute_step_2()          |
       - 44.11% -- | 35048.11 ms | optimize_lidar_odometry | lidar_odometry_utils_optimizers.cpp:1778, optimize_lidar_odometry() |
         - 0.25% - |   196.88 ms |             pre_hessian | lidar_odometry_utils_optimizers.cpp:1779, optimize_lidar_odometry() |
         - 33.18%  | 26366.49 ms |         hessian_compute | lidar_odometry_utils_optimizers.cpp:1813, optimize_lidar_odometry() |
         - 10.51%  |  8349.21 ms |            post_hessian | lidar_odometry_utils_optimizers.cpp:1891, optimize_lidar_odometry() |
     - 15.90% ---- | 12630.85 ms |        after_iterations | lidar_odometry_utils_optimizers.cpp:2560, compute_step_2()          |
       - 0.16% --- |   123.46 ms |           temp_save_laz | lidar_odometry_utils_optimizers.cpp:2591, compute_step_2()          |
       - 2.31% --- |  1835.14 ms |    propagate_trajectory | lidar_odometry_utils_optimizers.cpp:2630, compute_step_2()          |
       - 2.14% --- |  1696.83 ms | transform_points_global | lidar_odometry_utils_optimizers.cpp:2647, compute_step_2()          |
       - 11.24% -- |  8931.01 ms |              update_rgd | lidar_odometry_utils_optimizers.cpp:2657, compute_step_2()          |

Such profile could be usefull to understand the performance difference between windows and linux.

CMake configuration

• UTL_PROFILER_DISABLE option (default ON) — when ON, all profiler macros compile to no-ops with zero overhead
• To enable profiling: cmake -DUTL_PROFILER_DISABLE=OFF ...
• Dependent apps that don't need profiling have UTL_PROFILER_DISABLE always defined
• lidar_odometry_step_1 and pybind respect the CMake option via generator expression