drop arbitrary leansig.* workload; xmss.* already measures leanSpec PROD_CONFIG

`SIGTargetSumLifetime20W4NoOff` (V=39, W=4, TARGET_SUM=293, LIFETIME=2^20)
turned out to be a leanSig-internal benchmark sample — only used by
leanSig's own `src/bin/main.rs`, not pinned by any consensus client,
spec, or leanMultisig configuration. Our `xmss.*` workload already
measures `leansig_wrapper::LeanSigScheme`, whose parameters
(DIMENSION=46, BASE=8, TARGET_SUM=200, LOG_LIFETIME=32) exactly
mirror leanSpec's `PROD_CONFIG` per
`workspace/leanSpec/src/lean_spec/subspecs/xmss/constants.py`. So the
leansig.* group was a duplicate of nothing — measuring a configuration
no one ships.

Drop:
- The `leansig` workload module (workloads.rs).
- The Cli::Leansig{Keygen,Sign,Verify} variants + dispatch (main.rs).
- The three Workload entries in ALL_WORKLOADS (bench.py).
- The README "Workload groups" entry for leanSig.

Also drop `samples_override=300` on xmss.sign. The override was
statistically motivated when signing was randomized per-attempt
(rng-driven `rho`); cv ≈ 0.85 made n=30 give a ±30% CI on the mean.
devnet5's signing is fully derandomized — `rho` is a deterministic
PRF output of (sk.prf_key, epoch, message, attempts) — so for the
fixed (sk, epoch, message) the time_loop measures the same instance
each iteration and within-run cv collapses to CPU-clock noise. On
M1 Max devnet5: cv ≈ 0.014 at n=30, 95% CI on the mean ≈ ±0.5%.

The trade-off is that we now measure one specific (epoch, message)
instance rather than an average over the geometric retry distribution,
but that was already the case at n=300; the override only narrowed
the within-instance noise, not across-instance variation.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: unnawut

README.md

@@ -12,18 +12,15 @@ to Pages).
 
 ## Workload groups
 
-1. **leanSig** (variant `SIGTargetSumLifetime20W4NoOff`, lifetime 2^20) —
-   sign / verify
-2. **xmss** (the XMSS inside `leanEthereum/leanMultisig`, which is what
-   leanSpec actually consumes) — sign / verify at crate defaults
-3. **leanMultisig aggregation** at `LOG_INV_RATE_PROD=2` — flat
+1. **xmss** — leanSpec `PROD_CONFIG` GeneralizedXMSS (DIMENSION=46,
+   BASE=8, TARGET_SUM=200, LOG_LIFETIME=32), the scheme leanMultisig's
+   mainnet ships. sign / verify by default; opt-in keygen via
+   `--include-keygen`.
+2. **leanMultisig aggregation** at `LOG_INV_RATE_PROD=2` — flat
    aggregation over 125 / 250 / 500 / 1000 sigs, plus tree aggregation
    with fan-in 2 / 4 / 8 over 125 / 250 / 500-sig leaves. 13 aggregate
    variants total, all run by default.
 
-Key-generation is opt-in (`--include-keygen`) because lifetime-2^20 keygen
-adds significant wall time.
-
 ## Running on a target machine
 
 ### Prerequisites