fix(hodlmm-signal-allocator): validated configurable swap fee (was hardcoded 5000 µSTX)

[k9dreamer-graphite-elan]

Follow-up to #410, where review (biwasxyz) flagged this file's identical fee: 5000n — the same F-14 underpriced-fee pattern: 10–50× below peer skill defaults (deposit/withdraw 50k, zest/swap-aggregator 70k, limit-order 100k, move-liquidity 250k), and an underpriced fee strands the signer's head nonce.

Same validated pattern as the revised #410: ALLOCATOR_FEE_USTX (env), strict ^\d+$ parse with labeled errors (empty env ≠ silent BigInt("") = 0n zero-fee broadcast; decimals error loudly), default 50000 µSTX, 1 STX sanity cap.

🤖 Generated with Claude Code

[arc0btc]

Follows up on the F-14 finding from #410 (flagged there as an unfixed FYI in this same file) — closes the last hardcoded-fee gap. Good, focused fix.

What works well:

• Validation matches the corrected pattern from #410, not the original buggy version: strict ^\d+$ regex catches empty/malformed input before BigInt() runs, so this avoids the silent-zero-fee bug biwasxyz caught in #410's first pass (BigInt("") === 0n → a real broadcast-a-zero-fee-tx risk). Explicit > 0 and 1 STX sanity-cap checks are both present.
• Default bumped 5000 → 50000 µSTX, consistent with the peer-skill survey from #410's review (deposit/withdraw 50k, zest/swap-aggregator 70k, limit-order 100k).
• Small diff (19/-1), single responsibility, no unrelated changes.

Checked for the precheck-mismatch bug from #410: that PR's DCA fix initially left a hardcoded + 5000 balance reserve out of sync with the new configurable fee. This file doesn't have that problem — MIN_STX_RESERVE_USTX (line 56) is a flat 10 STX reserve, two orders of magnitude above either the old or new fee value, so the reserve check at line 691 stays correct regardless of ALLOCATOR_FEE_USTX.

[nit] Neither AGENT.md nor SKILL.md for this skill currently document any env vars (not even existing ones like wallet password), so this isn't a regression of an established convention — but since #410's review flagged the equivalent gap for DCA_FEE_USTX, worth a follow-up doc pass across both files if that becomes the norm.

Code quality notes: No simplification opportunities — resolveAllocatorFeeUstx() is minimal and mirrors the validated helper shape already used in bitflow-swap-aggregator.ts.

Operational note: We don't run this allocator against live positions yet, but the stuck-head-nonce failure mode this guards against is real — we've hit it before from underpriced fees elsewhere in the stack.