server: support live prefix rewind for GLM-5.3 and preserve common prefix on forks

[aaa2015]

Summary

In ds4-server, live session memory can often be reused when a new request shares a prefix with the active session. However, under high-intensity multi-turn agent sessions (such as tool-calling workflows), we observed frequent reason=token-mismatch cache misses and massive disk spills (e.g. live kv cache miss live=165755 prompt=4977 common=4973 reason=token-mismatch).

Investigation uncovered two root causes:

1. Engine capability gatekeeper lockout: ds4_server.c:12371 strictly checked ds4_engine_is_glm_dsa(s->engine), omitting GLM-5.3, which shares DS4_MODEL_FAMILY_GLM_DSA but is a distinct variant.
2. Strict equality requirement on fork: live_prefix_rewind_target() strictly asserted if (common != prompt_len) return -1;. When a multi-turn conversation or agent tool-call branched with common < prompt_len (e.g. 4,973 tokens matched out of 4,977, with 4 trailing new tokens), the entire resident session of 165K tokens was rejected, triggering an expensive disk cache eviction and a full prompt re-prefill.

Proposed Changes

1. ds4.h & ds4.c: Added bool ds4_engine_can_rewind(ds4_engine *e) to export whether the engine supports session rewinds, so the scheduler asks a question about rollback rather than about GLM.
2. ds4_server.c:
   • Generalized live_prefix_rewind_target(): when common < prompt_len, it rewinds to common, keeping the shared tokens resident and requiring ds4_session_sync() to only evaluate the suffix tokens. When common == prompt_len, it continues to rewind to prompt_len - 1 to re-evaluate the final token for logits generation.
   • Updated test_live_prefix_rewind_target() with regression cases covering fork rollbacks and matching the observed multi-turn token counts.

Note on Metal DSpark

An earlier revision of this PR admitted Metal DSpark through the new helper. That is removed here, because ds4_session_rewind() has no DSpark path yet:

• the rewind clears checkpoint_valid, so rewind_valid fails and the request falls back to a full rebuild — the exact cost the helper is meant to avoid;
• kv_cache_store_current(s, slot, "evict") then runs on the already-truncated session, so the persisted snapshot is shorter than the one that was resident.

Both are regressions relative to the current behaviour, where DSpark never enters the rewind branch. The DSpark branch is worth enabling together with the engine-side snapshot reuse that makes ds4_session_rewind() able to restore its state.

Verification

Machine: Apple M4, macOS 26.4.1, 16 GB. Backend: Metal.

make ds4_test && ./ds4_test --server
→ server: OK
→ ds4 tests: ok

make test-session-state
→ session state tests: ok
→ TP command tests: ok

make tests/test_session_state_gpu && ./tests/test_session_state_gpu
→ session state tests: ok

test_live_prefix_rewind_target() covers the fork-rollback cases added here,
including the observed multi-turn token counts (4973 of 4977 shared, 165755
resident). The model-backed suites (--logprob-vectors, --long-context, ...)
were not run because this machine has no GGUF checked out.