reset-readiness: greenfield deploy profile + markets write-through fix (#8)

[jdkajewski]

Makes packages/bot production-ready to run LIVE at the next greenfield weekly reset — deployment polish, not new earners. A readiness pass confirmed all earner subsystems are wired and home/expansion auto-detect; this PR removes the remaining reset-specific footguns and ships a reset-agnostic launch profile.

1. Fix issue #8 — markets write-through

persistence.putMarkets() sent the bot's Record<wp, Market> straight to the wire, but PUT /markets (BulkPutBody) expects Array<{ waypoint, data }> → 400, and the call is fire-and-forget so it was silently dropped (no market snapshot ever persisted → crawler/market history lost on restart).

• putMarkets now converts Record → Array<{waypoint, data}> before send.
• The GET side mismatched too (the api returns Array<{waypoint, data, updatedAt}>, the bot typed it as a Record) — getMarkets()/getMarket() now reduce the array back into the Record the cache expects.
• New contract test (persistence.markets.test.ts, 4 cases) validates the outgoing body exactly as BulkPutBody would, guards the old Record body as a regression, and round-trips GET so the bot↔api shapes can't drift again.

2. docker-compose — auto-detect home

• Dropped SYSTEM: ${SYSTEM:-X1-PP30} from the bot environment: block so a fresh reset auto-detects home via resolveHome() instead of pinning the previous reset's system. (environment: overrides env_file:, so a pin there would clobber a per-profile value.)
• Layered ./deploy/bot.greenfield.env as a second env_file: entry and documented the --env-file requirement for live runs.

3. Greenfield deploy profile

New deploy/bot.greenfield.env.example — a complete, reset-agnostic LIVE earner profile with no hardcoded system/waypoint/ship symbols:

• DRY_RUN=0, token=PASTE_AT_RESET, API_BASE_URL=http://api:3000, SYSTEM unset (auto-detect).
• Enables the self-provisioning earner stack: FLEET_SCALE, CONTRACTS, GATE_SUPPLY, MINE_FEED+MINE_EXPAND, GALAXY_CRAWL+AUTO_EXPAND — all of which auto-discover shipyards/tenders/outposts so no ship pins are needed. INPUT_FEED=0 (matches the battle-tested legacy profile).
• Floors mirrored from the legacy overnight_experiment profile (MIN_NET=1200, GATE_CREDIT_FLOOR=900000, FLEET_SCALE_FLOOR=30000, …) with reset-specific pins stripped and greenfield-safe values chosen.
• Issue-Value-weighted, lane-driven market scan budgeting (replace uniform 1:1 probe coverage) #2 scan-optimization levers (SCAN_*/COVERAGE_*/SCAN_BUDGET_*/FLEET_COVERAGE_*) left at default OFF — not needed to earn; enable + measure post-reset against the credits-per-request metric.
• Runbook header with the exact tomorrow flow: register → paste token → docker compose --env-file deploy/bot.greenfield.env up --build.

The legacy deploy/bot.env.example is untouched.

4. NEGOTIATOR default — honest gap signal ⚠

shared/config.ts defaulted NEGOTIATOR to a reset-specific ship 'SPACEJAM-DK-2-15' — a greenfield footgun (CONTRACTS=1 would dock/negotiate against a ship that doesn't exist on a fresh reset). Changed the default to '' and guarded the new-contract negotiate call on a non-empty NEGOTIATOR so an empty value cleanly skips auto-negotiation.

Gap: with NEGOTIATOR empty the bot still accepts offered/pending contracts but cannot negotiate NEW ones — the profile can't know the agent's command ship pre-reset. The runbook instructs setting NEGOTIATOR=<AGENT>-1 post-register for sustained contract income. A future enhancement could auto-pick the command ship; flagged for review.

Calibration assumption

The live .mjs history (live-metrics.json) is issue-#2 scan/lane data and not relevant to greenfield earner floors, which come from the battle-tested legacy overnight_experiment profile. Referenced in the env header as a comment for later scan-lever tuning.

Verification

• @st/shared build clean, @st/bot tsc --noEmit clean, eslint clean on changed files.
• 269 bot tests pass (incl. 4 new markets contract tests), 18 api tests pass (after the normal prisma generate codegen step).

Scope boundary: does NOT touch the parked issue-#2 optimization PRs (#9, #10).