Make the golden-runner quiet mode discoverable + measured status for #46

[pyrex41]

Why

Issue #46 has three complaints. Re-measuring all three on main @ 12fab4b:

1. Cold start ~1.5s — fixed by fasl: record shen.*lambdatable* deltas by name instead of serializing live lambdas #45/perf: urdr SHA workload series (fasl, hush-load, cons/fn/append, GC) #47. Warm trivial script is now 0.16 s (was 1.5 s).
2. urdr SHA suites ~12× CL — improved to 1.6–3.5× by perf: urdr SHA workload series (fasl, hush-load, cons/fn/append, GC) #47; the residual is structural.
3. -q unusable for golden runners — the fix shipped in perf: urdr SHA workload series (fasl, hush-load, cons/fn/append, GC) #47 as --hush-load, but it was documented only in source comments and bin/shen --help, had no regression test, and the README still showed -q as the only quiet mode. A runner author following the README lands straight back on the reported blocker (an empty run: on 41.2 *hush* gates pr itself).

This PR closes that last gap. No behaviour change to any existing mode.

Changes

• README — --hush-load in the launcher block, plus a Batch and golden-suite runners subsection: why -q structurally cannot work for this, a worked before/after, SHEN_HUSH_LOAD=1 for runners with fixed argv, and the cold/warm cache-independence guarantee.
• shen.boot{hush_load = true} — the embedder spelling, alongside jit = false. Applied before the idempotency guard: it is a pure output-mode flag with no boot work behind it, so an embedder handed an already-booted kernel can still ask for it.
• test/cli_spec.lua — 13 new assertions locking in all three modes (default / -q / --hush-load), that SHEN_HUSH_LOAD=1 is byte-identical to the flag, that --hush-load output is identical cold and on a warm fasl hit (so a golden file is cache-state independent), and the shen.boot{hush_load=true} path.
• doc/PERF-URDR-RESULTS.md — an independent re-run of issue Performance: cold start ~1.5s for a trivial script (~100x shen-cl); urdr SHA suites ~12x CL; -q unusable for golden runners #46's own repro, and an honest account of where the residual boot cost lives.

Measured (arm64 macOS, LuaJIT 2.1.1774638290, shen-cl 8df94be, urdr ba85b09)

Interleaved CL/Lua pairs, min-of-N, shen-lua warm + --hush-load.

Startup — (output "hi~%")

Configuration	wall
shen-lua warm (both caches)	0.16 s (median 0.18)
shen-lua first ever run (empty caches)	0.86 s
shen-lua SHEN_KERNEL_CACHE=off SHEN_FASL=off	0.71 s
shen-cl script	0.01 s

Warm boot CPU (min of 8): require 0.003 + load_kernel 0.051 + load_stdlib 0.100 = 0.154 s. The issue's initialise 1.77 s → 0.10 s.

load_stdlib is now the entire remaining cost, and it is 21 warm fasl hits (SHEN_FASL_DEBUG=1: 21/21 hit), not compilation. jit.p over it is flat — EQ 8%, append 6%, kdata_de 5%, is_cons / shen.assoc-> 5% each, fasl_read 3%. The cost is the replay rebuilding the environment, spread across the kernel's own list/assoc primitives; there is no hot spot left to cut. Closing the remaining ~16× to shen-cl means not rebuilding it at all (an image of the booted state — shen-cl's save-lisp-and-die equivalent), which is why this PR ships no boot micro-optimisation.

urdr suites (ALL PASS on both ports)

Suite	shen-cl	shen-lua	ratio	at issue open
shen/tests/prng	0.25 s	0.41 s	1.6×	~12×
shen/tests/search	0.80 s	2.52 s	3.1×	~12×
shen/tests/world	0.45 s	1.56 s	3.5×	—

Output modes, urdr prng

Invocation	stdout lines
shen-cl script run-tests.shen	232
bin/shen run-tests.shen (default)	239
bin/shen --hush-load run-tests.shen	71
bin/shen -q run-tests.shen	0

--hush-load yields exactly the 71 lines the suite itself prints; shen-cl's 232 are those same 71 plus its own 161-line load echo.

Testing

• luajit run-kernel-tests.lua — 134 passed / 0 failed, ok
• make test — 877 pass / 0 fail across 18 specs (was 864 before the new cases)
• luajit examples/openresty/selftest.lua — OK
• luajit examples/openresty-authz/selftest.lua — OK (file + lmdb + cosocket)
• luajit examples/envoy/selftest.lua — OK (ext_authz + Lua filter + upstream)
• urdr prng / search / world under bin/shen --hush-load — ALL PASS

Addresses item 3 of #46; #46 stays open for the structural suite-compute residual.

🤖 Generated with Claude Code