refine: remove the unused UnOp enum from the RQ IR

[prql-bot]

prqlc::ir::rq::UnOp has no producers or consumers. RQ models unary operations the same way it models every other operation — ExprKind::Operator { name: "std.neg", args }, lowered from PL in lowering.rs and translated in sql/gen_expr.rs — so nothing ever builds an rq::UnOp, matches on one, or folds one. The only UnOp the compiler actually uses is prqlc_parser's pr::UnOp, re-exported through ir::pl; the RQ copy carries a different variant set (Neg/Not against pr::UnOp's Neg/Not/Add/EqSelf), which makes the shared name an active confusion hazard for anyone reading both IRs.

This deletes the enum and drops it from the ir::rq re-export list. It is a public item of the prqlc crate, so this is technically a breaking change to that surface — but since no API in the crate accepts or returns the type, nothing downstream can be holding a value of it that any prqlc call would consume.

No regression test: the removal is a pure deletion of an unreferenced type, and cargo check -p prqlc --all-targets is the check that it was in fact unreferenced. Found by the nightly rolling survey.

How "unused" was established

A workspace-wide search for UnOp (and for un_op/unop, case-insensitive) returns only the parser's pr::UnOp definition in parser/pr/ops.rs and its uses — parser/expr.rs, parser/pr/expr.rs, ir/pl/mod.rs, semantic/ast_expand.rs, codegen/ast.rs — plus the two lines this PR removes. RqFold has no fold method for it, and no field of Expr or ExprKind references it, so it is unreachable from schema_for!(rq::RelationalQuery) and the generated RQ JSON schema is unchanged despite the JsonSchema derive. No committed schema or snapshot mentions it either.