optimizer runs faster and crashes less on large parameter sweeps

if you've run an optimizer sweep across a wide date range or a large number of parameter combinations, you may have seen it take an hour or crash entirely. this fixes two sources of unnecessary work that compounded on big sweeps.

first, the optimizer was building extra data for every single candidate tested — even though that detail is only needed for the small set of finalists that reach your results table. those candidates still get the full detail; the rest no longer pay for it.

second, market data was being fully copied into each background worker for every run. it's now handed off directly instead, which removes a large chunk of overhead on sweeps with many years of bars.

neither change affects what the optimizer tests or how results are scored — only the unnecessary work around each test is removed.