feat: tool_call_alias and depends_on for intra-turn tool scheduling

[Simon-Free]

Summary

Expose two scheduling hints to the LLM on every tool schema, and strip them before they reach tool handlers:

• tool_call_alias: string - optional alias the model can use to refer to a tool call by a short name later in the same turn.
• depends_on: string[] - list of prior tool_call_ids or aliases. The model uses this to express sequential dependencies between tools that it wants executed one after the other rather than in parallel.

Also coerces string-typed params (sent by LLMs that flatten JSON) into their schema-declared types: "42" ? 42 for an integer property, "true" ? True for boolean, '[{...}]' ? list/dict for array/object.

What's in scope here

This PR only injects the schema fields and performs param coercion + stripping. Runtime enforcement of depends_on ordering in the agent loop is not yet implemented - the model gets a hint and can call tools in order manually, but the registry does not re-order parallel executions based on depends_on. Deferred to a follow-up PR so this one stays small and review-friendly.

Changes

File	+/-	What
tool_registry.py	+86	_SCHEDULING_PROPS constant, _coerce_params split into per-type coercers dispatched through _COERCERS, wrapper over get_tool_schemas that injects scheduling props, wrapper over execute_tool that strips scheduling props and coerces types
tests/test_tool_scheduling.py	+104	Unit coverage of schema injection, coercion per type, pass-through on invalid input, and stripping at dispatch
tests/test_tool_scheduling_e2e.py	+96	Two e2e scenarios via agent.run + mocked stream: (1) every schema the LLM sees carries the scheduling props, (2) when the LLM emits a tool_call that includes tool_call_alias + depends_on, those are gone by the time the tool handler runs

Cleanups folded in

• _coerce_params' silent except (ValueError, json.JSONDecodeError): pass replaced by a dispatch table where each coercer explicitly returns the original value on failure, with a comment explaining the intent ("tool handler reports the real type mismatch").
• Fix test_scheduling_params_stripped which called execute_tool(name, params) without the required config arg - it was failing since the branch landed.

Ref #43

[Simon-Free]

Changes in this updatenn### Removed: type coercion (split to separate PR)nCoercion is an independent concern -- moved to its own PR with bug fixes.nn### Added: ID uniqueness enforcementnPorted id_uniquify.py from bouzecode. Without this, when the LLM reuses IDs like r1 across turns, you get duplicate tool_call_id errors from the API.nn### Fixed: input_schema-style schema injectionnScheduling props (tool_call_alias, depends_on) were injected at the wrong level for Anthropic-style schemas. Now correctly targets input_schema.properties.nn### Testsn- 7 new ID uniquify unit testsn- 1 new e2e test (ID reuse across turns)n- Removed coercion tests (moved to coercion PR)

[Simon-Free]

Changes in this update

Removed: type coercion (split to separate PR #62)

Coercion is an independent concern - moved to its own PR with bug fixes.

Added: ID uniqueness enforcement

Ported id_uniquify.py from bouzecode. Without this, when the LLM reuses IDs like r1 across turns, you get duplicate tool_call_id errors from the API.

Fixed: input_schema-style schema injection

Scheduling props were injected at the wrong level for Anthropic-style schemas. Now correctly targets input_schema.properties.

Tests

• 7 new ID uniquify unit tests
• 1 new e2e test (ID reuse across turns)
• Removed coercion tests (moved to PR feat: type coercion for string-typed tool params #62)

[Simon-Free]

Actually depends on #47 (needs conftest.py)

[Simon-Free]

Dependency: This PR depends on #47 (pr3-checkpoint-stderr-tokens). Please merge #47 first.nnThe tests/conftest.py shared fixtures (_no_quota) are introduced in #47. This PR extends conftest with scripted_stream + receiver_tool for scheduling-specific e2e tests.