Add scoped .instructions.md files for source code and tests

[dgenio]

Why

GitHub Copilot supports path-scoped instruction files (.github/instructions/*.instructions.md) that activate only when Copilot is working on files matching a glob pattern. This allows more targeted guidance than the global copilot-instructions.md:

• Source code gets production-specific rules (Pydantic patterns, exception hierarchy, etc.)
• Test code gets testing-specific rules (fixture patterns, assertion style, etc.) — and can exclude Copilot Code Review from seeing test-framework boilerplate rules that would generate false-positive review comments.

Scope / Proposed changes

Create two files in .github/instructions/:

1. python-source.instructions.md — scoped to chainweaver/**/*.py
2. testing.instructions.md — scoped to tests/**/*.py, with excludeAgent: "code-review"

Proposed contents

File 1: .github/instructions/python-source.instructions.md

---
applyTo: "chainweaver/**/*.py"
---

# Python source instructions — ChainWeaver

These instructions apply when editing production source code in `chainweaver/`.

## Module conventions
- Every module starts with `from __future__ import annotations`
- Every module has a module-level docstring explaining its role
- All public classes and functions have Google-style docstrings

## Pydantic patterns
- Use `pydantic.BaseModel` for all data models (v2 API)
- Use `model_dump()` not `.dict()` (v2 migration)
- Use `model_validate()` not `.parse_obj()` (v2 migration)
- Field descriptions via `Field(description="...")` for schema clarity

## Exception patterns
- All custom exceptions inherit from `ChainWeaverError`
- Exception constructors accept context kwargs: `tool_name`, `step_index`, `detail`
- Import exceptions from `chainweaver.exceptions`, never define inline

## Export rules
- Any new public class/function MUST be added to `chainweaver/__init__.py` `__all__`
- If adding a new module, add its imports to `__init__.py`

## Executor invariant
- `executor.py` must remain deterministic: no LLM calls, no network I/O, no randomness
- The executor runs a compiled graph — it does not make decisions

File 2: .github/instructions/testing.instructions.md

---
applyTo: "tests/**/*.py"
excludeAgent: "code-review"
---

# Testing instructions — ChainWeaver

These instructions apply when editing test files. They are excluded from 
Copilot Code Review to avoid false-positive review comments on test patterns.

## Framework
- Use `pytest` exclusively (no `unittest.TestCase`)
- Use plain `assert` statements (pytest rewrites them for clear diffs)

## Structure
- File naming: `tests/test_<module>.py`
- Group related tests in classes: `class TestSuccessfulExecution:`
- One logical assertion per test when practical

## Fixtures
- Use `@pytest.fixture()` with explicit parentheses
- Fixtures return fully configured objects (tools, flows, executors)
- Prefer fixture composition over deep setup logic
- Scope fixtures appropriately (`scope="function"` is default and preferred)

## Coverage patterns
- Every new feature needs: happy path + at least one error/edge case
- Test exception messages, not just exception types
- For executor tests: verify both `ExecutionResult.success` and `StepRecord` contents

## Anti-patterns in tests
- Do NOT mock internal ChainWeaver classes unless testing integration boundaries
- Do NOT use `time.sleep()` — tests must be deterministic and fast
- Do NOT import from `chainweaver` using relative paths

Why excludeAgent: "code-review" on testing instructions?

When Copilot Code Review processes a PR that includes test files, it would receive testing-specific instructions like "use @pytest.fixture() with explicit parentheses" and may flag existing test code that doesn't follow these conventions — even if the PR didn't modify those tests. This creates noisy, false-positive review comments.

By setting excludeAgent: "code-review", these instructions guide code generation (Copilot Chat, Copilot Coding Agent) but are hidden from Code Review, reducing noise.

Labels to apply

• documentation
• priority:P2
• ai-friendly
• milestone:v0.1.0

Depends on

• Add .github/copilot-instructions.md with coding conventions and validation commands #53 — .github/copilot-instructions.md should exist first (scoped instructions complement, not replace, the global file)

Acceptance criteria

• .github/instructions/python-source.instructions.md exists with applyTo: "chainweaver/**/*.py"
• .github/instructions/testing.instructions.md exists with applyTo: "tests/**/*.py" and excludeAgent: "code-review"
• Both files use valid YAML frontmatter syntax
• No content from global copilot-instructions.md is duplicated — these files add path-specific rules only
• Pydantic v2 API patterns cited are correct (e.g., model_dump() not .dict())
• Exception hierarchy matches actual code in chainweaver/exceptions.py

Avoid drift/duplication notes

• Global conventions (Ruff config, validation commands, PR rules) stay in .github/copilot-instructions.md — NOT repeated here.
• Architecture context stays in AGENTS.md — NOT repeated here.
• These files contain only path-specific rules that would be noise in the global file.
• If a scoped rule becomes universal, move it to copilot-instructions.md and remove from here.

References

• GitHub Docs: Creating instruction files
• applyTo and excludeAgent syntax
• Add .github/copilot-instructions.md with coding conventions and validation commands #53 — Global copilot-instructions.md
• Add AGENTS.md with architecture map, decision context, and agent guardrails #51 — AGENTS.md