[iamalwaysuncomfortable]

The Python rebalance tools currently turn every funding budget into an exact quote, so they cannot use the flexible recovery and execution-time liquidity sizing added by amm-v3 #110. This adds the corresponding behavior from Veil #220, preserving existing exact calls.

• Add exact-amount, exact-liquidity, and amount-driven planners and execution helpers. Sync and async planners share the existing pure math; liquidity execution remains on the sync client.
• Accept token caps/minimums and fixed funding/refunds. Amount-driven plans retain a zero contract target and report estimated_liquidity separately. Flexible surplus remains collectible on the successor.
• Submit prebuilt plans verbatim, reject conflicting overrides, and validate u128 values, token bounds, funding/refund conflicts, and exact accounting before preparing a transaction.
• Expose the modes through agent/MCP tools with precise decimal-string amounts, and update the README and generated agent guides.

Validation:

• Full offline Shield Swap suite: 572 passed, 8 skipped, 95 deselected.
• Generated-context freshness and size checks, sync/async surface checks, and git diff --check passed.
• Pyright 1.1.414 reports the same 15 existing diagnostics on the unchanged base and this branch, with no new diagnostics.
• Independent review found no actionable issues. A final regression test also verifies that agent tools reject fractional/boolean amounts rather than silently converting them.

Reference heads: Veil 3d5ef1d65e1f1efa818cd2d7e7dab5bda5a3da17; amm-v3 6676393c723ba5d3caad688dde20731c8896128c.

No funded/live transactions were run. Flexible calls require a compatible contract deployment; deployment compatibility has not been verified live. Contract/ABI and shared liquidity arithmetic are unchanged.