feat: apply pep574 out-of-band pickling to DefaultContainer

[larme]

What does this PR address?

This PR is an improvement of #3606 . Instead of using pickle5 to avoid memory copying of a single numpy ndarray, codes improved by this PR can avoid copying of any composite data structure (e.g. a list of ndarray) containing numpy ndarray or pandas dataframe.

Fixes #(issue)

Before submitting:

• Does the Pull Request follow Conventional Commits specification naming? Here are GitHub's
  guide on how to create a pull request.
• Does the code follow BentoML's code style, both make format and make lint script have passed (instructions)?
• Did you read through contribution guidelines and follow development guidelines?
• Did your changes require updates to the documentation? Have you updated
  those accordingly? Here are documentation guidelines and tips on writting docs.
• Did you write tests to cover your changes?

[codecov]

Codecov Report

Merging #3736 (537442f) into main (4a6d332) will not change coverage.
The diff coverage is 0.00%.

@@          Coverage Diff          @@
##            main   #3736   +/-   ##
=====================================
  Coverage   0.00%   0.00%           
=====================================
  Files        148     149    +1     
  Lines      12408   12447   +39     
=====================================
- Misses     12408   12447   +39     

Impacted Files	Coverage Δ
src/bentoml/_internal/runner/container.py	0.00% <0.00%> (ø)
src/bentoml/_internal/utils/pickle.py	0.00% <0.00%> (ø)

[bojiang]

Potential:

• pandas DataFrame doing
• binary file (need investigation, not now)
• PIL.Image (not supported naturally)

[bojiang]

Will continue to review after dataframe supported

[aarnphm]

@larme should we also mention here the performance cost for this?