refactor(inference-grpc,PIECE-8): delete hardcoded worker-count ceilings + magic constants

[joelteply]

Summary

CBAR-PIECE-8 (vhsm-d1f4 audit pass 1, surfaced again in #1316 ALPHA-GAP's 'Concrete deletion target' callout): get_num_workers() in inference-grpc/main.rs had three anti-patterns that violate the dynamic / broker-owned-concurrency rule. All three deleted.

Anti-patterns deleted

	Was	Why deleted
1	clamp(1, 8) ceiling on the env-var path	A Blackwell with 128GB RAM capped to 8 workers, same as a MacBook Air. Supervisor's value must pass through verbatim.
2	clamp(1, 4) ceiling + magic 2GB-per-worker constant on autodetect	The 2GB number is wrong for every model that isn't a 7B Q4_K_M. Hardcoded ceiling silently capped throughput on big hardware.
3	Silent Default: 2 workers fallback when sys-info fails	The exact "guess and degrade" anti-pattern vhsm-d1f4 audit pass 1 + 6 explicitly called out.

Also deleted: ~/.continuum/config.env file reading (static-config-file violates dynamic rule); sys-info crate dep (only consumed by the deleted auto-detect path).

New resolve_num_workers()

1. INFERENCE_WORKERS env var — the channel a supervising continuum-core sets at process spawn (broker-derived). Value passes through verbatim. No clamping. Supervisor knows the live hardware + memory pressure; this binary doesn't second-guess.

2. Env var unset → num_cpus::get_physical().max(1). Hardware-derived, never zero, one info log so operator sees the fallback. Documents that continuum-core supervisor SHOULD set INFERENCE_WORKERS based on its PressureBroker lease — the broker integration is the next PR in this chain.

3. INFERENCE_WORKERS=0 or invalid → Err with bad value named. main() propagates to abort startup. No silent default. Surfaces the config bug at the source instead of launching with a dead pool.

Test plan

14 passing on cargo test --no-default-features -- --test-threads=1:

• env var passes through verbatim (8)
• env var=64 not capped (pins no-ceiling guarantee; would have been clamped to 8 before)
• env var=0 → Err with 0 in message
• env var=not-a-number → Err with value named (operator sees what was set)
• env var unset → num_cpus::get_physical() fallback (matches host)
• env var empty (INFERENCE_WORKERS=) → Err (empty ≠ unset; user meant to set something)
• env var=1 (lower boundary) → passes
• env var=-1 (negative, shell underflow case) → Err

Note: env-mutating tests must run serial (--test-threads=1). Pinned in the with_env helper docstring + the test module name makes it discoverable.

What this enables (CBAR-SUBSTRATE alignment)

One less hardcoded ceiling between the supervisor's PressureBroker and the actual inference pool size. Once a future PR wires continuum-core to spawn inference-grpc with INFERENCE_WORKERS=<broker-lease>, the concurrency budget is dynamic + supervisor-controlled end-to-end. The deletion landed here unblocks that wiring without further refactoring of inference-grpc.

Closes one of the three deletion targets listed in #1316 ALPHA-GAP's 'Concrete deletion target' callout.

Coordination

Two prior attempts to edit this file (per local stash history) tripped multi-tab races + got flagged by vhsm-d1f4 as keeping env-var static-config reflex. This PR keeps the env var ONLY as the supervisor-set channel; deletes the file-reading + magic-constant scaffolding. Should land cleanly because the scope is narrow + non-conflicting with concurrent docs PRs.