fix: evaluate CURRENT_TIMESTAMP at runtime, not compile time

[0xhsn]

Changelog

• CURRENT_TIMESTAMP, CURRENT_DATE, and CURRENT_TIME are now evaluated at execution time via Insn::Function instead of being baked into static String8 instructions at compile time
• Added now() as a function alias for datetime(), matching PostgreSQL behavior
• Removed unused use crate::functions::datetime import from expr.rs

Context

When using pgmicro as the database backend for a web application, INSERT statements with DEFAULT CURRENT_TIMESTAMP columns and prepared queries referencing CURRENT_TIMESTAMP returned stale values from statement preparation time rather than execution time.

The root cause: translate_literal() in core/translate/expr.rs called exec_datetime_full(&[]) during compilation and emitted the result as a constant string. The datetime runtime functions already handle zero arguments correctly by calling set_to_current() — the fix is simply to emit Insn::Function calls instead of pre-computing the value.

Testing and Deployment

6 new sqltests added to testing/sqltests/tests/scalar-functions-datetime.sqltest. All 8650 passing sqltests still pass (8 pre-existing hash-join mismatches unrelated).

[glommer]

Sorry it took me a while to get to this =(
I have just merged a new release of Turso into pgmicro, and fixed some edges. now() and the other functions are working in the master branch now. But the CURRENT_TIMESTAMP issue is still not resolved. Do you want to update the PR with it ?

[0xhsn]

@glommer No worries! Rebased on master, kept just the CURRENT_TIMESTAMP/CURRENT_DATE/CURRENT_TIME runtime-eval fix in core/translate/expr.rs (now emits Insn::Function instead of compile-time Insn::String8), and dropped the now() alias since the PG translator already handles it.