[Misc] Refactor aclgraph accuracy test to use logprob-based comparison (#7455)

### What this PR does / why we need it?

Replace text-match assertions with a two-tier logprob accuracy check:

- Prefill (token 0): assert token ID is identical between eager baseline
and compiled mode, then verify logprob matches within `atol`.
- Decode (tokens 1-2): if chosen tokens match, compare logprobs
directly; if they differ, cross-lookup the baseline token in the
compiled model's top-20 distribution and assert the assigned logprob is
within `decode_atol` (defaults to 2x atol). This tolerates minor argmax
drift caused by floating-point differences while still catching
distribution divergence.

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.17.0
- vLLM main:
8a680463fa

---------

Signed-off-by: wangli <wangli858794774@gmail.com>

.github/workflows/scripts/config.yaml

@@ -5,6 +5,8 @@ e2e-singlecard:
   estimated_time: 69
 - name: tests/e2e/singlecard/test_auto_fit_max_mode_len.py
   estimated_time: 70
+- name: tests/e2e/singlecard/test_eager_mode_acc.py
+  estimated_time: 255
 - name: tests/e2e/singlecard/test_aclgraph_accuracy.py
   estimated_time: 839
 - name: tests/e2e/singlecard/test_aclgraph_batch_invariant.py