[async] Demote struct-fors in async compilation

[k-ye]

1. I separated compile_to_offloads() into two functions: compile_to_offloads() and compile_to_executable(). The former stops right before struct-for demotion. This allows me to run async mpm99 without any patches now.
2. Note that after converting a struct-for to a range-for, we need to call irpass::fix_root_block_kernel() to fix the block kernels.

Related issue = #1541

[Click here for the format server]

---

[codecov]

Codecov Report

Merging #1593 into master will increase coverage by 0.11%.
The diff coverage is 46.66%.

@@            Coverage Diff             @@
##           master    #1593      +/-   ##
==========================================
+ Coverage   67.47%   67.58%   +0.11%     
==========================================
  Files          40       40              
  Lines        5624     5634      +10     
  Branches      982      983       +1     
==========================================
+ Hits         3795     3808      +13     
+ Misses       1661     1657       -4     
- Partials      168      169       +1     

Impacted Files	Coverage Δ
python/taichi/main.py	42.35% <0.00%> (ø)
python/taichi/lang/shell.py	51.38% <47.45%> (+12.67%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0f4ce50...dbb6d87. Read the comment docs.

[yuanming-hu]

Thank you so much! Everything here looks great to me.

[archibate]

Cool!

[xumingkuan]

LGTM!