fix: drop prompt rules that reference disabled tools

[JulianFriedle]

Summary

_assemble_prompt() filters TOOL_SECTIONS through included = tool_names - disabled, but appends the static rule blocks (_AGENT_RULES, _API_AGENT_RULES, _DOMAIN_RULES) unchanged. A disabled tool therefore loses its schema and its description while keeping its behavioural rule, so the prompt can instruct the model to use a tool it has not been given. On dev @3b6c16916233 the manage_memory rule sits in both _AGENT_RULES (L437) and _API_AGENT_RULES (L451), so the prompted and the native-tool-calling path are equally affected.
This adds _strip_disabled_tool_rules(), which removes bullet lines naming a disabled tool in backticks, applied to the finished prompt at both return points of _assemble_prompt(). It is a no-op when nothing is disabled, so default behaviour is unchanged.

Target branch

• This PR targets dev, not main.

Linked Issue

Fixes #6285

Type of Change

• Bug fix (non-breaking — fixes a confirmed issue)

Checklist

• I searched open issues and open PRs — this is not a duplicate.
• This PR targets dev
• My changes are limited to the scope described above — no unrelated refactors or whitespace changes mixed in.
• I actually ran the app (docker compose up) and verified the change works end-to-end.
• I did not run the app/runtime validation.

How to Test

1. Build and start

docker compose up -d --build

2. Verify the disabled tool no longer leaves its rule behind

docker compose exec odysseus python -c "
from src.agent_loop import _assemble_prompt
p = _assemble_prompt({'manage_memory','manage_notes'}, {'manage_memory'})
print('manage_memory in prompt:', 'manage_memory' in p)
print('manage_notes  in prompt:', 'manage_notes' in p)
"

Expected: False / True. Only the disabled tool's rule is removed; the rule
for the enabled tool stays. On dev without this change the same command
prints True / True.

3. Unit tests

python3 -m pytest tests/test_disabled_tool_rules.py -q

Expected: 6 passed.

4. End-to-end

Put a tool with a behavioural rule into disabled_tools, send a message
matching that rule's trigger in agent mode, and confirm the model now uses an
available tool instead of making no call at all.

Visual / UI changes — REQUIRED if you touched anything that renders

Not applicable — this PR touches src/agent_loop.py and adds a test file. No
rendering code, no CSS, no static/js/.

• I am not an LLM agent submitting a bulk PR.