feat: subagent personalities

[tlongwell-block]

Add built-in investigator and planner subagent personalities

Subagents can now use two specialist modes without defining custom subrecipes:
• investigator - Read-only research into codebases and systems
• planner - Strategic decomposition before complex tasks

Additional personalities can be added later by placing recipes in crates/goose/src/prompts/subrecipes/

Usage: subrecipe: "investigator" or subrecipe: "planner" with instructions passed via the instructions parameter.

Implementation:

• Extends prompt_template.rs to serve bundled subrecipes from existing CORE_PROMPTS_DIR
• Adds investigator.yaml and planner.yaml to src/prompts/subrecipes/
• Extracts params_to_values() and combine_instructions() helpers to eliminate duplication in subagent_tool.rs
• Adds max_turns field to Recipe struct/builder for turn limits in subrecipes

Why this approach:
Reuses existing prompt infrastructure rather than creating parallel systems. Bundled subrecipes are discovered via iter_bundled_subrecipes() and built with the same build_recipe_from_template() path as user-defined subrecipes.

Builds on the work in #5893

[tlongwell-block]

Probably should reconsider this in the context of recipes/skills later

[DOsinga]

yes. maybe the two of us can have a little chat about this and figure out the diverse dimensions of skills, recipes, claude plugins, apps and agents.md include hierarchies and come up with a plan?