Bug fixes, AI chat improvements and critical security fix (key filtering)

What's Changed

Bug fixes

• feat(annual): warn when a solar kWp figure looks like Watts was entered by @springfall2008 in #4860
• GE Cloud: surface write rejections instead of counting them as successes by @springfall2008 in #4902
• fix(fetch): make the cloud/load divergence model actually modulate by @chalfontchubby in #4871
• fix(security): stop apps.yaml downloads and debug dumps leaking credentials by @springfall2008 in #4910
• fix(octopus): review follow-ups for #4849, #4850 and #4852 by @mgazza in #4873

AI Chat

• Add regex and start/end time bounds to the get_log tool by @springfall2008 in #4855
• Report the context length Ollama will actually give, not the model's ceiling by @springfall2008 in #4859

Docs

• docs: add Hanchu iESS mid-window charge/discharge time update automation by @upton68 in #4903

Tests

• fix(test): assert the ML training invariant instead of peak RSS by @chalfontchubby in #4912

Dev

• Triage bot: have BOT_CLEANUP resolve merge conflicts against main by @springfall2008 in #4882
• Triage bot: disclose bot authorship on every comment it posts by @springfall2008 in #4863
• Triage bot: --ollama/--ollama_review model selection, auto-review opened PRs by @springfall2008 in #4861

Full Changelog: v8.54.2...v8.54.3