Fix AUTO_INCREMENT zero handling and tests

[robll-v1]

What type of PR is this?

• API-change
• [ ✅] BUG
• Improvement
• Documentation
• Feature
• Test and CI
• Code Refactoring

Which issue(s) this PR fixes:

issue #14490

What this PR does / why we need it:

Fixed MatrixOne’s AUTO_INCREMENT behavior to match MySQL: when sql_mode does not include NO_AUTO_VALUE_ON_ZERO, inserting 0 into an AUTO_INCREMENT column is treated as NULL and allocates a new id. This prevents HyBench’s “Duplicate entry '0'” errors.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[XuPeng-SH]

The UPDATEsemantics may be changed: the current logic converts 0 to NULLwhenever HasAutoColis true (including in the UPDATEprocess). This means that UPDATE t SET auto_col=0 will allocate a new auto-increment value when sql_modedoes not contain NO_AUTO_VALUE_ON_ZERO, which is inconsistent with MySQL (where UPDATEtypically treats 0 as a normal value).

[XuPeng-SH]

Each batch reads sql_modeand performs a string operation (ToUpper+ Contains). This incurs a small amount of string allocation and CPU overhead, consider caching this boolean value at the session level to reduce the cost.