docs: compile TROUBLESHOOTING.md with CUDA and drone errors (closes #63)

[DsThakurRawat]

Description

Introduces docs/TROUBLESHOOTING.md to help users quickly resolve common deployment and configuration errors at the edge, in the cloud, and on robots/drones.

Key Additions:

• CUDA & GPU Errors: Fixes for libcudnn version mismatches, insufficient NVIDIA drivers, and GPU OOM during graph capture.
• Jetson-Specific Issues: Solutions for JetPack versioning and thermal throttling during continuous inference.
• ROS2 Bridge Issues: Debugging steps for missing state data and missing ROS2 (rclpy) installations.
• Drone & MAVROS Issues: Specific fixes for MAVROS FCU connections, missing IMU data, and action output routing.
• Export & Validation: Handling opset incompatibilities and 'Not yet verified' statuses.
• Quick Diagnostics: A handy cheat sheet of commands (nvidia-smi, reflex doctor, etc.).

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Infrastructure / CI update

How Has This Been Tested?

• pytest tests/ passes locally
• reflex doctor sanity check
• Other (please specify): Verified formatting and cross-referenced with codebase/schema.

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have kept the PR scoped to a single concern (one concern per PR)

[rylinjames]

Thanks @DsThakurRawat — the error-signature coverage in this PR (libcudnn mismatch, CUDA OOM, JetPack version, MAVROS FCU connection, opset rollback, the whole catalog) is real and useful. Credit preserved via Co-Authored-By on the commit in the replacement PR.

Same lighter-touch approach I took with #130: this is content that needed grounding more than rewriting. The replacement only refreshes:

• onnxruntime-gpu==1.18.0 → >=1.25.1 (current floor since v0.9.2's Blackwell support bump)
• cuDNN floor: 9.0 → 9.5
• Driver floor: R525 → R555 (for cuDNN 9.5+, per the v0.9.4 doctor guard)
• Added Blackwell sm_120 section (RTX 5090 / B200 — bit a real customer for 2 weeks, now caught by v0.9.3 doctor guard)
• Added "First step: run reflex doctor" with the four v0.9.4 guards (multi-GPU arch, Jetson R35, cuDNN/driver skew, TRT EP empirical session test)
• Replaced your reference to PR fix(ros2): support IMU state vector for drone embodiments #121's auto-detection with the actual shipped --state-msg-type flag from fix(ros2): pluggable state extractors (joint_state / imu / odom) #133 (joint_state / imu / odom dispatch — your #121 was superseded by it)
• Fixed cross-refs to verification.md (renamed from understanding_verification.md in docs: VERIFICATION.md trust-receipt guide (closes #35) #136) and adding_a_robot.md (shipped in docs: Adding-a-Robot embodiment cookbook (closes #64) #135)
• Renamed the file to troubleshooting.md for consistency with sibling docs

All your other error patterns and fix commands kept verbatim — they're correct and battle-tested. Closing in favor of the replacement PR. Please keep contributing!