fix: report subscriptions actually created, not a flag-derived guess

[fcostaoliveira]

Fixes #43.

TotalSubscriptions was total_channels * subscribers_per_channel, but -subscribers-per-channel does not drive subscription creation. Real fan-out is -clients × channels-per-connection, where channels-per-connection comes from -min/-max-number-channels-per-subscriber (both default 1). The reported number and the real one share no terms.

The failure mode is the awkward kind: a caller reads -subscribers-per-channel 10 as ten subscribers per channel — the help text says exactly that — and the output then confirms the belief with a figure derived from the same inert flag. Nothing signals the request went unmet.

Changes

1. Count subscriptions as they're created, and report that. The old value was derived from inputs, so it could never detect a mismatch between request and reality.
2. Warn when an explicit -subscribers-per-channel goes unmet, naming the flags that actually control fan-out. Gated on the flag being explicitly set (via flag.Visit) — the default differing from the real count is normal and not the caller's doing, so warning there would just be noise on every run.
3. Fail loudly on non-dense placement in subscribe modes. Only dense is implemented and the branch had no else, so an unrecognised placement silently created zero subscribers and produced a clean zero-message result — which reads as "the server delivered nothing" rather than "we subscribed to nothing".

total_messages is recomputed from the corrected count, so the -messages termination target no longer derives from the unmet request.

Verified against a live server

case	warning	TotalSubscriptions	actual
explicit -subscribers-per-channel 10, unmet	✅ fires	50	50
correct fan-out via channels-per-subscriber	none	500	500
plain default run	none	20	20
-subscribers-placement-per-channel sparse	exits 1 with reason	—	(was silently 0)

Before this change, case 1 reported 1000.

Downstream

This is live in the benchmark spec: the ...-50K-subscribers-5k-conns suite passes exactly the case-1 arguments, so it generates the same 5000 subscriptions as its 5000-subscriber sibling. Fixed separately in redis/redis-benchmarks-specification#500 — that YAML fix and this one are independent and each correct on its own.

I deliberately did not make -subscribers-per-channel control fan-out (option 1 in #43), since that changes behaviour for anyone currently passing it. Happy to do that instead if you'd prefer; this PR takes the option that only makes the reporting truthful.

[github-actions]

🐳 Docker Build Validation

✅ Docker build successful!

Platforms tested:

• ✅ linux/amd64 (built and tested)
• ✅ linux/arm64 (build validated)

Git SHA: 0cb58923fc771d178e0d0b0f5be8f5f80d3fea3e

Docker Hub Status: ✅ Docker Hub credentials configured

Image details:

• Single platform: pubsub-sub-bench-pr:pr-44
• Multi-platform: pubsub-sub-bench-pr:pr-44-multiplatform

Tests performed:

• ✅ Docker Hub credentials check
• ✅ Help command execution
• ✅ Version output validation
• ✅ Multi-platform build validation

The Docker image is ready for deployment! 🚀