feat: add ThoughtProof reasoning evaluator example for ACP jobs

[ThoughtProof]

Summary

Adds an example showing how to use ThoughtProof as a reasoning-based evaluator for ACP jobs via the on_evaluate callback.

What it does

Checks whether a job deliverable is well-reasoned before accepting payment — adversarial multi-model critique returns ALLOW or HOLD.

Why

Complements #193 (Maiat trust evaluator):

• Maiat checks trust — is this provider historically reliable?
• ThoughtProof checks reasoning — is this specific deliverable well-justified?

Both evaluators together: reliable provider + sound reasoning = full coverage.

Usage

from thoughtproof_evaluator_example import thoughtproof_evaluator

acp_client = VirtualsACP(
    ...,
    on_evaluate=thoughtproof_evaluator(stake_level="medium", domain="general"),
)

## Payment

x402, $0.02-$0.05 USDC on Base per evaluation (stake-proportional). Use [purl](https://purl.dev) for testing

cc @JhiNResH

[ThoughtProof]

Updated this branch to align with pot-sdk v2.0 (now live at api.thoughtproof.ai).

Changes:

• Verdict enum: ALLOW / BLOCK / UNCERTAIN (was ALLOW / HOLD / DISSENT)
• HOLD → BLOCK in all logic branches
• UNCERTAIN is now a safe escalation state (not "model confusion")
• Updated docstrings and expected test outputs

HOLD and DISSENT are now internal calibration states only — they no longer appear in public API responses.

───

Example — ThoughtProof as ACP job evaluator:

acp_client = VirtualsACP(
...,
on_evaluate=thoughtproof_evaluator(stake_level="medium", domain="general"),
)

Deliverable: "ETH trade thesis with sentiment only, no data"

ThoughtProof → BLOCK — job.evaluate(False), payment withheld

Deliverable: "Structured analysis with price levels, stop-loss, and rationale"

ThoughtProof → ALLOW (confidence: 0.84) — job.evaluate(True), payment released

Deliverable: "Ambiguous output — incomplete reasoning"

ThoughtProof → UNCERTAIN — accepts with review flag

The evaluator hook sits between job completion and payment release. ThoughtProof verifies the reasoning quality of the deliverable before job.evaluate(True) is called.