fix: apply redirect correctly when resolving wildcard on this

[kgutwin]

The following PRQL fails compilation:

from albums
select { album_id, title, artist_id }
sort this.*

With the error:

Error: 
   ╭─[ :1:1 ]
   │
 1 │ ╭─▶ from albums
   ┆ ┆   
 3 │ ├─▶ sort this.*
   │ │                 
   │ ╰───────────────── table instance cannot be referenced directly
   │     
   │     Help: column name might be missing?
───╯

This turns out to be because the process for resolving the wildcard this.* ends up returning the incorrect tuple {albums = {this.albums.album_id, this.albums.title, this.albums.artist_id}} as seen in the debug log below:

[DEBUG prqlc::semantic::resolver::expr] resolving ident this.`*`...
[TRACE prqlc::semantic::module] lookup: this._self
[TRACE prqlc::semantic::module] lookup: _self
[TRACE prqlc::semantic::module] ... following redirect albums
[TRACE prqlc::semantic::module] lookup: _self
[TRACE prqlc::semantic::module] ... result of redirect albums: {["albums", "_self"]}
[TRACE prqlc::semantic::module] ... following redirect this
[TRACE prqlc::semantic::module] lookup: this._self
[TRACE prqlc::semantic::module] ... following redirect albums
[TRACE prqlc::semantic::module] lookup: this._self
[TRACE prqlc::semantic::module] ... result of redirect albums: {}
[TRACE prqlc::semantic::module] ... result of redirect this: {}
[TRACE prqlc::semantic::module] ... following redirect that
[TRACE prqlc::semantic::module] lookup: this._self
[TRACE prqlc::semantic::module] ... result of redirect that: {}
[TRACE prqlc::semantic::module] ... following redirect _param
[TRACE prqlc::semantic::module] lookup: this._self
[TRACE prqlc::semantic::module] ... result of redirect _param: {}
[TRACE prqlc::semantic::module] ... following redirect std
[TRACE prqlc::semantic::module] lookup: this._self
[TRACE prqlc::semantic::module] ... result of redirect std: {}
[DEBUG prqlc::semantic::resolver::expr] ... resolved to _wildcard_match
[DEBUG prqlc::semantic::resolver::expr] ... which is Expr: {albums = {this.albums.album_id, this.albums.title, this.albums.artist_id}}

The tuple instead should be {this.albums.album_id, this.albums.title, this.albums.artist_id}.

The underlying cause of this issue is in this section of resolve_ident_wildcard():

prql/prqlc/prqlc/src/semantic/resolver/names.rs

Lines 221 to 228 in 6ffb2b9

	let mut res = self.root_mod.module.lookup(&ident_self);
	if res.contains(&ident_self) {
	res = HashSet::from_iter([ident_self]);
	}
	if res.len() != 1 {
	return Err(format!("Unknown relation {ident}"));
	}
	let module_fq_self = res.into_iter().next().unwrap();

When ident_self is ["this", "_self"], self.root_mod.module.lookup(&ident_self) returns a HashSet containing both ident_self and the redirect (["this", "albums", "_self"]). However, the next if res.contains() statement drops the redirect, so the value of module_fq_self ends up just being ["this", "_self"]. This ends up getting wildcard-expanded to the tuple representing the literal module this._self, rather than the correct redirect this.albums._self.

This PR makes two small changes:

• After successfully following a redirect in Module::lookup(), we remove the ident that triggered the redirect from the result set. This avoids inadvertently triggering "ambiguous name" results.
• The resolve_ident_wildcard() function has been refactored to use the same match decls.len() structure as used in other resolve_ident functions. This harmonizes error conditions across the name resolve process and avoids the bug behavior triggered by the current if res.contains() logic.

[prql-bot]

CI is failing on queries::results::wildcard_this because the new integration test is missing the integration__queries__results__wildcard_this.snap snapshot — task prqlc:test-all (or task prqlc:pull-request) accepts the snapshot locally, but the file then has to be committed alongside the others.

Project guidance (prqlc/prqlc/tests/CLAUDE.md) actually prefers small inline insta::assert_snapshot! tests in prqlc/prqlc/tests/integration/sql.rs over .prql integration tests: each .prql file generates ~6 snapshots, and for a compilation-stage fix like this you can get equivalent coverage with one snapshot. There's a near-twin pattern at tests/integration/sql.rs:6020 (test_select_bare_wildcard). Replacing the new files with a single test_sort_this_wildcard would also sidestep the missing-results-snapshot issue entirely.

Substantively the fix reads correctly to me — the Module::lookup change drops the literal match from res only when a redirect actually resolves the same ident non-empty, and the only place that ambiguity arises in practice is _self lookups (since columns live in per-input sub-modules, not directly under this). The resolve_ident_wildcard cleanup harmonizes nicely with resolve_ident_core/resolve_ident_fallback.