fix(pytest): surface diagnostic context when no tests are collected

[YOMXXX]

Summary

`rtk pytest` (no args) collapsed pytest's collection output to a single terse line — `Pytest: No tests collected` — whenever the summary parser couldn't find a `passed/failed/skipped` count. Reporter (#1417) saw this on a real project where `rtk pytest tests/test_cli_dispatch.py -v` worked but `rtk pytest` did not, with no way to tell whether it was a cwd issue, a pytest config issue, or an RTK bug.

Two real situations both hit this:

1. No tests in cwd — pytest emits `no tests ran in X.XXs` instead of a counted summary line. The user was left guessing why nothing was found.
2. Collection error — `ImportError` in a conftest, missing fixture import, etc. pytest emits `errors during collection` + an `ERROR` block, then `no tests ran`. The actionable error was being thrown away.

Fix

Add `diagnose_no_tests`: when the existing summary collapses to `No tests collected`, scan the raw pytest output for the lines that explain why — `rootdir:`, `configfile:`, `testpaths:`, `collected N items`, any `ERROR …` line, plus the `no tests ran` / `no tests collected` / `errors during collection` markers — and emit them verbatim under a header so the LLM and the user can act on them.

```
Pytest: no tests collected
═══════════════════════════════════════
rootdir: /Users/x/proj
configfile: pyproject.toml
testpaths: tests, integration
collected 0 items
no tests ran in 0.05s
```

Capped at 15 lines (a tee-style overflow hint would be wrong here — diagnostic lines are bounded in practice).

Falls back to the terse default when no recognisable marker is present (e.g. RTK's parser broke but tests actually ran), so we don't flood the output with arbitrary noise on a parser regression.

Behaviour

Scenario	Before	After
Empty cwd, `rtk pytest`	`Pytest: No tests collected`	+ rootdir / configfile / testpaths / "no tests ran"
Collection `ImportError`	`Pytest: No tests collected`	+ the actual `ERROR collecting …` line
Quiet-only `no tests ran in 0.05s` (no rootdir present)	terse	terse (markers fence keeps it from over-firing)
Tests pass	unchanged	unchanged
Tests fail	unchanged	unchanged

Test plan

• `test_filter_no_tests_verbose_surfaces_rootdir_and_configfile` — rootdir / configfile / testpaths / collected count / "no tests ran" all reach the consumer.
• `test_filter_no_tests_collection_error_surfaced` — ImportError-during-collection scenario shows the `ERROR` line.
• `test_filter_no_tests_quiet_mode_falls_back_to_terse` — bare `no tests ran in 0.05s` (no rootdir to anchor on) keeps the original short verdict.
• Pre-existing `test_filter_pytest_no_tests` updated to assert the new enriched output while keeping the same verdict header for grep-ability.
• `cargo fmt --all` clean.
• `cargo clippy --all-targets` zero warnings.
• `cargo test --bin rtk -- --test-threads=8` → 1912 passed (was 1909).

Notes

• The original reporter's underlying confusion (was the cwd wrong? was it pytest config? was it RTK?) is now answerable straight from the rtk output: rootdir tells you where pytest looked, configfile tells you which config it used, and testpaths tells you which subtree it scanned.
• Doesn't touch the `-q` defaults or the runner's cwd — those are still pytest's standard behaviour.

Fixes #1417