[HPU] Add lazy mode back

[yiliu30]

Background: previous PRs forced the use of compile mode for HPU, but lazy mode can still work, albeit slowly, and some INC UTs rely on lazy mode.

This PR:

• check environment variables for compile mode if needed

# lazy mode
PT_HPU_LAZY_MODE=1 python xxx

# compile mode
PT_HPU_LAZY_MODE=0 PT_ENABLE_INT64_SUPPORT=0 python xxx

• Add UTs for both lazy mode and compile mode