fix(pm): compute avg_resolution_time running mean in uint128

[web3blind]

Summary

Small follow-up to P5 (oracle-metrics) on pm (#124): the running-mean formula for pm_oracle_object::avg_resolution_time accumulates in uint64, but avg_resolution_time (uint32) * (n-1) + rt can reach the uint64 boundary once n approaches the uint32 max — the product of two uint32 values sits within a hair of 2^64, and adding rt can tip it over.

Change

libraries/chain/pm_evaluator.cpp, in pm_resolve_market_evaluator::do_apply:

const uint64_t sum =
    ((fc::uint128_t((uint64_t)ora.avg_resolution_time) * (uint64_t)(n - 1)
      + fc::uint128_t(rt)) / n).lo;
ora.avg_resolution_time = (uint32_t)sum;

fc::uint128_t is already used elsewhere in the same file (time-penalty math), so no new includes.

Verification

• Formula equivalence checked against the original integer math on boundary cases (avg=0..UINT32_MAX, n=1..4e9, rt=0..UINT32_MAX) — matches on all cases, and confirms the original uint64 path lands within 1 bit of overflow on the extreme (bit_length()==64).
• git diff --check clean; single file, 6 insertions / 2 deletions.
• Full C++ build not run here (no local Boost toolchain); change is a pure arithmetic widening with identical semantics.