[enhancement](load) avoid schema copy to reduce cpu usage

[zhannngchen]

Proposed changes

Issue Number: close #xxx

Problem summary

some user reports that the cpu usage is quite high when data loading.
The flame graph shows that copy of the schema cost lots of cpu resources.

Checklist(Required)

1. Does it affect the original behavior:
   • Yes
   • No
   • I don't know
2. Has unit tests been added:
   • Yes
   • No
   • No Need
3. Has document been added or modified:
   • Yes
   • No
   • No Need
4. Does it need to update dependencies:
   • Yes
   • No
5. Are there any changes that cannot be rolled back:
   • Yes (If Yes, please explain WHY)
   • No

Further comments

If this is a relatively large or complex change, kick off the discussion at dev@doris.apache.org by explaining why you chose the solution you did and what alternatives you considered, etc...

[github-actions]

clang-tidy made some suggestions

[hello-stephen]

TeamCity pipeline, clickbench performance test result:
the sum of best hot time: 36.76 seconds
load time: 509 seconds
storage size: 17123237211 Bytes
https://doris-community-test-1308700295.cos.ap-hongkong.myqcloud.com/tmp/20230119030741_clickbench_pr_83804.html

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[github-actions]

clang-tidy review says "All clean, LGTM! 👍"

[zhannngchen]

_schema is inited from request.schema(), and it's life cycle is longer than delta writers, so we can use _schema directly to avoid copying the schema twice(the first time is here, the second time is in the ctor of DeltaWriter)

[dataroaring]

LGTM

[github-actions]

PR approved by at least one committer and no changes requested.

[github-actions]

PR approved by anyone and no changes requested.