Signed-off-by: idellzheng <idellzheng@tencent.com>

The blend test's decoder has been failing with a 400s port-8200
timeout since LMCache#2994 landed (Apr 18). The actual error hidden in the
decoder log is:

    TypeError: LMCacheMPWorkerAdapter.__init__() got
    multiple values for argument 'mq_timeout'

Root cause: LMCache#2994 simplified `LMCacheMPSchedulerAdapter.__init__` and
`LMCacheMPWorkerAdapter.__init__` to take positional args
`(..., vllm_block_size, parallel_strategy, mq_timeout, ...)`, removing
the backward-compat `(world_size, kv_rank, vllm_block_size, tp_size,
parallel_strategy=None)` form. The vllm-bundled caller at
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_mp_connector.py
line 148 was not updated -- it still passes the legacy shape:

    LMCacheMPWorkerAdapter(
        server_url, zmq_context, model_name,
        world_size, kv_rank, vllm_block_size,
        mq_timeout=mq_timeout,
        heartbeat_interval=heartbeat_interval,
    )

After LMCache#2994, positional arg 6 (cache_config.block_size) lands on the
`mq_timeout` parameter, and then the explicit `mq_timeout=` kwarg
collides -- hence "multiple values". The decoder crashes during
EngineCore init and never binds its HTTP port, which surfaces as the
generic `wait_for_server` 400s timeout up in run-blend-test.sh.

Because `lmcache_mp_connector.py` is imported lazily and the actual
class is resolved via a try/except that prefers
`lmcache.integration.vllm.vllm_multi_process_adapter` (this file) over
the vllm-bundled fallback, an installed LMCache wheel with the new
signature guarantees the crash on every recent vllm nightly. This is
why every branch forked from dev since Apr 21 20:58 has failed blend
with a mysterious decoder timeout.

Fix: keep the new internal convention (the class stores
`parallel_strategy` and derives world_size/tp_size from it) but
accept BOTH calling conventions at the __init__ boundary. A small
`_coerce_parallel_strategy()` helper inspects arg types at call
time:

- If the second polymorphic slot is a `ParallelStrategy`, use the new
  (vllm_block_size, parallel_strategy) form directly.
- Otherwise treat the three slots as the legacy
  (world_size, kv_rank, vllm_block_size) and synthesize a
  minimal `ParallelStrategy` from them plus the `tp_size` kwarg.

Callers currently in-tree (`lmcache_mp_connector_0180.py`) already
use the new form, so they keep working. The vllm-bundled caller's
legacy positional shape now also works without changes, unblocking
the blend test decoder.

Verified locally on the K3s host with the tensormesh/cacheblend
image and the PR's source:
- Before fix: decoder EngineCore crash on init, port 8200 never
  binds, run-blend-test.sh fails at the wait_for_server step.
- After fix: prefiller ready in 43s, decoder ready in 16s,
  shuffle_doc_qa benchmark completes, "[PASS] Blend integration
  test completed successfully" in the build log.

Signed-off-by: Samuel Shen <slshen@tensormesh.ai>

LMCache#2994 simplified `LMCacheMPSchedulerAdapter.__init__` and
`LMCacheMPWorkerAdapter.__init__` signatures, removing the legacy
(world_size, kv_rank, vllm_block_size, tp_size, parallel_strategy=None)
params. The vllm-bundled caller at
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_mp_connector.py:148
still passes the legacy shape (world_size, kv_rank, vllm_block_size
positionally + mq_timeout as a kwarg). After LMCache#2994, positional arg 6
lands on the new `mq_timeout` param and the kwarg collides:

    TypeError: LMCacheMPWorkerAdapter.__init__() got
    multiple values for argument 'mq_timeout'

The decoder crashes during EngineCore init, never binds port 8200,
and surfaces as the generic 400s wait_for_server timeout in
run-blend-test.sh. Every feature branch forked from dev since
Apr 21 20:58 has failed blend this way.

Fix: restore the pre-LMCache#2994 signature with legacy ints + optional
`parallel_strategy`, and synthesise a minimal ParallelStrategy from
world_size/kv_rank/tp_size when it isn't passed. Everything
downstream that reads `self.parallel_strategy` keeps working
unchanged.

Verified locally on the K3s host with the tensormesh/cacheblend
image: prefiller ready in 43s, decoder ready in 16s, shuffle_doc_qa
benchmark completes, "[PASS] Blend integration test completed
successfully" in the build log.

Signed-off-by: Samuel Shen <slshen@tensormesh.ai>