Release v3.0.1: NanoGPT local mode auth and XML streaming fixes

- Fix NanoGPT authentication in local mode by passing API key to Docker container
- Populate authVars from environment variables in local mode
- Harden XML stream processor to prevent incomplete <PlandexFinish tags in output
- Add stop sequence variants and split-chunk detection for finish tags
- Tighten commit message XML prompt to enforce exact <commitMsg> element
- Include NanoGPT subscription-only model pack (GLM-4.6 + DeepSeek R1)

This release resolves 402 authentication errors in local mode and eliminates dangling XML tags in streamed responses.

Release v3.0.0 - GLM-4.6 Support & NanoGPT Subscription Optimization

Major Features:
- GLM-4.6 model support (200K context, superior coding performance)
- GLM-4.6-thinking variant for complex reasoning tasks
- Optimized NanoGPT subscription model pack
- Fixed chat-only context bloat issue

New Models:
- zhipu/glm-4.6 - Advanced agentic and coding capabilities (z-ai/glm-4.6 on NanoGPT)
- zhipu/glm-4.6-thinking - Reasoning version with visible thinking (z-ai/glm-4.6:thinking on NanoGPT)

Improvements:
- Chat-only requests no longer include heavy project maps
- Comprehensive NanoGPT subscription model documentation
- Optimal model pack for subscription users
- Correct model naming for NanoGPT provider

Model Pack Configuration:
- Planner/Architect/Coder: GLM-4.6 (best performance)
- Builder: GLM-4.6-thinking (complex reasoning)
- Summarizer: Claude 3.5 Haiku (fast/cheap)
- Names/Commits: GPT-4.1-mini (simple tasks)

All models confirmed working on NanoGPT subscription at zero additional cost.

Release v2.2.2: Update repository URLs and fix install script

Server v2.2.3 - NanoGPT provider selection debug logging

Server v2.2.2

🐞 Bug Fixes:
- Fix for panic in execTellPlan when active plan is nil
- Resolved 'Invalid memory address or nil pointer dereference' panic during plan execution
- Added nil check in panic recovery defer block to prevent race condition crashes
- Improved server stability and error handling in tell execution flow

chore: bump CLI release version to 2.2.9

feat: enhance NanoGPT setup with subscription vs balance mode options

- Add interactive billing mode selection during NanoGPT setup
- Explain subscription mode (fixed monthly cost) vs balance mode (pay-per-use)
- Provide guidance on switching modes with CLI commands
- Update provider description to mention billing options
- Improve user experience for choosing appropriate billing model
- Reference existing connect-nanogpt/disconnect-nanogpt commands

Add NanoGPT subscription mode support

- Added UseNanoGPTSubscription preference in OrgUserConfig
- Created separate provider for NanoGPT subscription endpoint
- Added provider filtering logic to choose correct endpoint based on user preference
- Added CLI commands: nanogpt, connect-nanogpt, disconnect-nanogpt
- Enables users to switch between balance and subscription billing modes
- Fixes balance depletion issue for NanoGPT subscription users

Enhanced installation and upgrade process

- Install script now updates existing local server installations
- Upgrade process now updates both CLI and local Docker server
- Added 'plandex update-server' command for manual server updates
- Ensures CLI and server versions stay in sync
- Fixes issue where server runs old code after CLI upgrade

Key improvements:
- Installation prompts to update existing server directory
- Automatic server update during CLI upgrades
- Standalone server update command
- Better error handling and user feedback
- Maintains Docker container state during updates

Fix nil pointer dereference in model fallback handling

- Fix missing BaseModelConfig in FallbackResult when no fallback is needed
- Add safety check in client_stream.go to prevent nil pointer access
- Resolves panic: 'Invalid memory address or nil pointer dereference'

This fixes the runtime error that occurred when the model execution
pipeline tried to access BaseModelConfig.HasClaudeMaxAuth on a nil pointer.