feat: add flashsac offpolicy integration

[TATP-233]

Summary

• add algo=flashsac to the unified off-policy training entrypoint
• add a dedicated src/unilab/algos/torch/flash_sac/ package with runner, learner, network, layer, and update modules
• add Hydra config/schema coverage for flashsac and task=flashsac/g1_sac/{mujoco,motrix}
• extend actor/collector wiring and tests for FlashSAC compose and learner updates

Validation

• make check
• make test

Impact Scope

• offpolicy training entrypoint
• torch offpolicy runtime wiring
• Hydra structured configs and task owners
• FlashSAC algorithm implementation and unit coverage

[TATP-233]

Merged into #240 by cherry-picking 542ea1e onto fix/issue-236-time-limit-bootstrap; continuing review there.