docs: refresh stale scale figures across the documentation - #1039

Supersedes #774. Credit to @telagod for the original audit — the premise was right and worth doing; the numbers had simply aged out, so everything here was re-derived from source at main (5f1cade0) rather than carried forward.

#774 proposed 6,300+ tests against an actual 7,499, and ~2.7 MB for a binary that measures 4.66 MB. Both are corrected below.

Scale figures

Provenance for each, all confirmed to run as written:

• Providers — src/providers/factory.zig is the actual single source of truth for OpenAI-compatible providers. Counted as grep -cE '\.\{ \.name = "' src/providers/factory.zig → 110. The 10 core implementations are the real provider files, excluding composition wrappers (router.zig, reliable.zig) and shared helpers.
• Tools — grep -rhoE 'pub const tool_name = "[a-z0-9_]+"' src/tools/ | wc -l → 40.
• Channels — one struct per channel under src/channels/, excluding the *_ingress, *_api, *_presenter, outbox, dispatch, and external_protocol helpers → 24. The architecture tables listed a partial set that silently omitted Teams, Max, WeChat/WeCom, and Weixin; those are now listed and the row is labelled with the count. English and Chinese updated together.
• Files/LOC — git ls-files cross-checked against find; all three methods agree on 293 / 289,734.

Binary size — a real discrepancy, not just a stale number

README.md and CLAUDE.md claimed 678 KB, and AGENTS.md §2.2 set a sub-1 MB ReleaseSmall target. A host build measures:

$ zig build -Doptimize=ReleaseSmall && ls -l zig-out/bin/nullclaw
-rwxr-xr-x  4889528  zig-out/bin/nullclaw        # ~4.66 MB, aarch64-macOS

That is ~7x the long-quoted figure and ~4.7x the stated target. All three documents now state the measured value, AGENTS.md marks the sub-1 MB goal as an open gap rather than a met constraint, and the verification command is included so the number is re-checkable instead of asserted.

I'd treat this as a product-accuracy decision rather than a docs fix: either the binary gets smaller, or the target gets restated honestly. That's a separate conversation and I have not attempted either here — this PR only makes the documentation tell the truth about the current state.

Keeping the numbers current

The root cause of #774 going stale is that these figures are hand-maintained and nothing checks them. AGENTS.md §1 now carries the exact derivation commands, and a repo-wide sweep confirms no remaining 5,640 / 5,300 / 678 KB / 0.15.x references in the touched files (AGENTS.md §7.6 treats a version-pin mismatch as a hard fail, so that was checked explicitly).

For reviewers: the "last verified at 5f1cade0" marker is the point at which these stop being true. Please re-derive rather than increment.

Validation

• zig build test --summary all — 13/13 steps, 7490/7499 passed, 9 skipped, 0 failures, 0 leaks
• zig build -Doptimize=ReleaseSmall — exit 0
• zig fmt --check src/ — exit 0
• Docs only, 5 files, +42/−25. No code touched.

Not included

Deliberately split out to keep this reviewable: the CLAUDE.md deduplication from #775 (a 1-file architectural change that also needs the stale Zig 0.15.2 pin corrected). Separate PR follows.