Detect Assign effects

[lionel-]

Progress towards #1338
Branched from #1341

Adds Assign as the fourth effect kind, after NSE, attach, and source. A call or operator that binds a name with Assign now creates a real definition that feeds the use-def map, exports(), goto, and rename exactly like a syntactic <-. These Assign effects are shadow-, source-, and attach-aware.

Two forms:

• Calls like assign("x", value) and delayedAssign("x", value). The name is the string first argument, the value the next positional. Bails when envir = or pos = is set, since then the binding lands outside the current scope and we can't state it as a fact.
• Operators like x %<>% f() (magrittr), x %<~% expr (rlang), x := expr (S7).

Impl notes:

• A new DefinitionKind::Assign models this effect in the semantic index.

• A new RangedAstPtr wrapper of AstPtr allows carrying the trimmed range of the name node. AstPtr only carries the full range, and requires the root node to resolve a trimmed range, which is unergonomic.

Goto and rename now work with custom Assign calls. When a name is bound in string form (e.g. assign("x", ...)) the rename occurs in place as a string to keep the quotes. To make this easier I've dropped the normalisation of strings during renames (which matches your preference Davis, cf previous discussion).

To support that cleanly, the decision of how a renamed name is spelled moved out of the LSP layer into oak_ide. This way the LSP layer only converts ranges and wires files.

Positron Release Notes

New Features

• Code navigation and renaming now understands custom assignment calls and operators like assign(), delayedAssign(), %<>% from magrittr, := from S7, etc.

Bug Fixes

• N/A