refactor(any): Remove lifetime parameter from AnyArguments

[iamjpotts]

Does your PR solve an issue?

Similar to #3957, allows more flexibility with passing references as bind parameters to query! macros when using the Any driver.

This is the first of two prs, after which Query::bind and Query::try_bind argument lifetimes will be relaxed to be consistent across the drivers. The second pr adds a test which is a compile-time check of the new lifetime allowances for arguments. That test will not compile with just this pr.

Second of the two prs:
#3960

Is this a breaking change?

Yes:

• removes lifetime parameters from structs that are a part of the public api for the Any driver
• changes enum variants for AnyValueKind

[joeydewaal]

Just as a note:
To implement something like #875, you'd have to defer the encoding of the arguments until your in the driver. One way to do that is to store the arguments in a Vec<&dyn Encode>. Which is something we could do now since we have a lifetime in ArgumentBuffer.

(There are pro's and con's ofcourse, it'd save a copy here and here but all bind calls would require a reference.)

[iamjpotts]

Just as a note: To implement something like #875, you'd have to defer the encoding of the arguments until your in the driver. One way to do that is to store the arguments in a Vec<&dyn Encode>. Which is something we could do now since we have a lifetime in ArgumentBuffer.

(There are pro's and con's ofcourse, it'd save a copy here and here but all bind calls would require a reference.)

One of the design choices of the crate is for the query! and query_file! macros to have semantics similar to println! and format! - meaning you can provide a variable name or expression as a parameter, but not have to explicitly borrow it. The implementation of the query! and query_file! macros always borrows the parameters.

That works with the postgres flavor in sqlx 0.8 because its implementation of Arguments<'_> does not have a lifetime (and thus has the static lifetime). It does not currently work with all of the drivers in 0.8.x, since some of them implement Arguments<'_> with a struct having a lifetime parameter.

When a driver's Arguments implementation has a lifetime, that lifetime is the 'q lifetime on Query and its bind method (used by the code generated by the macros). For postgres, 'q == 'static.

sqlx/sqlx-core/src/query.rs

Lines 75 to 86 in bab1b02

	impl<'q, DB: Database> Query<'q, DB, <DB as Database>::Arguments<'q>> {
	/// Bind a value for use with this SQL query.
	///
	/// If the number of times this is called does not match the number of bind parameters that
	/// appear in the query (`?` for most SQL flavors, `$1 .. $N` for Postgres) then an error
	/// will be returned when this query is executed.
	///
	/// There is no validation that the value is of the type expected by the query. Most SQL
	/// flavors will perform type coercion (Postgres will return a database error).
	///
	/// If encoding the value fails, the error is stored and later surfaced when executing the query.
	pub fn bind<T: 'q + Encode<'q, DB> + Type<DB>>(mut self, value: T) -> Self {

So the opposite is actually the case for named parameter support - to preserve println! similar usage, instead of borrowing the parameter with something like Vec<&dyn Encode>, the parameter has to be either a) immediately encoded into an owned driver-specific buffer for drivers supporting named parameters, or b) the parameter value has to be copied, cloned, or otherwise turned into an owned value with a 'static lifetime.