feat(dialect): report whether placeholders are positional

[iamralch]

Closes #3.

Dialect described placeholder syntax but not the consequence that follows
from it: whether a bind can be referenced more than once.

That does not matter to this crate — transpile binds each literal once and
never points at an earlier one — but it decides correctness for a caller
splicing a fragment that names one bind from two places.
sqlx-aip's key-set cursor predicate
pins each more-significant column in every clause after the first:

-- numbered: $1 is bound once, referenced twice
("title" > $1) OR ("title" = $1 AND "id" > $2)
-- positional: each ? consumes its own bind, so the value list repeats
("title" > ?)  OR ("title" = ?  AND "id" > ?)

Two ordering fields bind two values on Postgres and three on SQLite for the same
predicate. Getting it wrong raises nothing — the binds shift by one and the page
silently resumes from the wrong row.

The change

fn is_positional(&self) -> bool {
    self.placeholder(1) == self.placeholder(2)
}

Defaulted, so no existing Dialect impl breaks, and the default is exactly the
inference sqlx-aip was making from outside. It handles the awkward middle case:
a dialect emitting SQLite's numbered ?1 / ?2 form is positional in syntax but
still addressable, renders the two differently, and is correctly reported as
not positional.

The three built-ins answer directly rather than rendering anything, and the
blanket impl Dialect for &D forwards it — a missed forward there would look
like a silent fallback to the default rather than a compile error, so a test
calls through a reference.

One test pins the default against the three explicit answers. Without it nothing
would notice the inference drifting away from what the built-ins say, and the
inference is what every non-overriding custom dialect gets.

Testing

58 lib tests (+3), clippy and fmt clean across sqlite,mysql.