## Implementation Plan

### Problem
`_render_with_playwright` and the `crawl` top-level function in the DeepSearch crawler use an incorrect Playwright context manager pattern: they call `__aenter__` on the `async_playwright()` result but then attempt `__aexit__` on the **Playwright object** (the result of `__aenter__`) instead of on the **context manager** returned by `async_playwright()`. This causes an `AttributeError: 'Playwright' object has no attribute '__aexit__'` when Playwright is actually launched and then cleaned up.

### Files to Modify
1. **`devplacepy/services/jobs/deepsearch/crawl.py`** (or whatever file contains both functions — confirm absolute path via `git grep` before editing; assume `/workspace/repo/devplacepy/services/jobs/deepsearch/crawl.py` based on investigation line references).

### Changes

#### 1. `_render_with_playwright` function – `own_browser=True` branch (around line 175)

**Before:**
```python
pw = await async_playwright().__aenter__()
try:
    browser = await pw.chromium.launch(...)
    ...
finally:
    await pw.__aexit__(None, None, None)
```

**After (use `async with`):**
```python
async with async_playwright() as pw:
    browser = await pw.chromium.launch(...)
    ...
```
No manual `__aenter__`/`__aexit__` needed; the `async with` block handles it correctly.

#### 2. `_render_with_playwright` function – shared browser case (already correct – `browser` passed in, no context manager needed)

No change required.

#### 3. `crawl` function (around line 294) – top-level browser launch and management

**Before (conceptual):**
```python
pw = await async_playwright().__aenter__()
try:
    browser = await pw.chromium.launch(...)
    ...
finally:
    await pw.__aexit__(None, None, None)
```

**After (use a separate variable for the manager):**
```python
pw_manager = async_playwright()
pw = await pw_manager.__aenter__()
try:
    browser = await pw.chromium.launch(...)
    ...
finally:
    await pw_manager.__aexit__(None, None, None)
```
This preserves the two-variable pattern needed because the browser lifecycle spans the whole function (cannot simply use `async with` since `browser` is passed to helper methods).

(If the function already uses a `try/except` that catches launch failures, keep that; the fix only changes the `finally` cleanup target.)

### Verification Steps

1. **Run unit tests**  
   `pip install -e '.[dev]' -q && make test-unit`  
   All existing tests (including those that mock `fetch_page`) must pass. No new failures introduced.

2. **Run linter**  
   `pip install -q ruff && ruff check .`  
   Zero lint errors on the changed files (and ideally no new lint issues overall).

3. **Manual sanity check** (optional but recommended)  
   - Confirm the fix compiles: `python -c "import ast; ast.parse(open('devplacepy/services/jobs/deepsearch/crawl.py').read())"`  
   - If possible, run a deepsearch job that triggers the Playwright fallback to verify no `__aexit__` error (not strictly required for “Definition of Done” but good to confirm).

### Definition of Done
- [ ] The two changes above are applied to the exact file(s) identified by `git grep`.  
- [ ] `make test-unit` runs to completion with zero failures (pre-existing failures in `test_compounds.py` and `test_owns_instance_true_for_creator` are unrelated and expected to remain).  
- [ ] `ruff check .` reports no errors on the modified file(s).  
- [ ] No new warnings or errors during the test run.