[lsp] Accept legacy boolean diagnostic tag support

[purefunctor]

Fixes #552.

Legacy clients can send a boolean for textDocument.publishDiagnostics.tagSupport. Since the switch to gen-lsp-types, that makes the workspace reject initialize and the protocol actor silently discard client capabilities.

Ignore this unused boolean field before either actor decodes capabilities. Leave object-form tag support unchanged. This restores initialization compatibility without introducing a second capability model or separate normalization paths.

The regression test runs both actors and covers true, false, and the object form. It also checks UTF-8 negotiation and the outgoing workspace/configuration request so accepting initialization cannot hide a fallback to default capabilities.

Verification

• Reproduced InvalidParams with the regression test before the fix.
• cargo check -p iris-lsp-server -p iris-lsp --tests passed.
• cargo nextest run -p iris-lsp-server -p iris-lsp-workspace -p iris-lsp: 89 passed.
• After the tuple-destructuring review change, cargo check -p iris-lsp --tests and all 4 iris-lsp tests passed.
• just format and git diff --check passed.
• just licenses completed with license-discovery warnings and left THIRDPARTY.toml unchanged.

Amp thread: https://ampcode.com/threads/T-01a0fb04-e03a-7329-9c0d-fa7571238921

[claude]

Review against AGENTS.md: no issues found.

The protocol actor is the only place that receives initialize before handing it to both decoders: Session::negotiate, which falls back to default capabilities on a decode error, and the workspace's InitializeParams decode, which rejects with InvalidParams. So rewriting the boolean there fixes it in the layer that owns it, with a single normalization path. Nothing in the server reads diagnostic tag support (grep finds no DiagnosticTag/tag_support uses), so dropping the legacy boolean loses no behavior. The regression test drives both actors through the existing iris-lsp session harness. It checks UTF-8 negotiation and the workspace/configuration request, so a fallback to default capabilities would fail it. That fits the test-ownership rules.

Checks run:

• cargo check -p iris-lsp-server -p iris-lsp --tests: passed
• cargo nextest run -p iris-lsp-server -p iris-lsp: 41 passed, 0 failed
• just format --check: clean
• just licenses: only license-confidence warnings; git status clean afterwards, so THIRDPARTY.toml is unchanged
• No integration categories are affected, so I didn't run just t; no .snap.new files are present

Not verified: I didn't revert the fix locally to confirm the new test fails without it. The PR description reports reproducing InvalidParams before the fix.

[purefunctor]

Granite Review

Council

Role	Mode	Status	Attempt	Amp thread
Council member	review-gpt-6-luna	completed	1	T-01a0fb10-379b-72c8-958c-39fef28824fd
Council member	review-gpt-6-sol	completed	1	T-01a0fb10-36f6-725f-a317-d77860c52414
Synthesis	review-astra-synthesis	completed	1	T-01a0fb15-0241-7328-9e0e-a5609c51a965

Findings

No actionable findings.

[github-actions]

Compatibility regression report

Package set 81.1.0 for PureScript 0.15.15.

✅ The candidate introduces no compatibility errors.

Diagnostic class	Base	Candidate	Introduced	Fixed
Compiler errors	0	0	0	0
Compiler warnings	36	36	0	0
Verifier errors	0	0	0	0

Introduced errors

None.

Fixed errors (0)

None.

Warning changes (0 introduced, 0 fixed)

Introduced

None.

Fixed

None.

Candidate errors (0)

None.

Candidate warnings (36)

• deno@0.0.5/src/Deno.purs:38:17 — CustomWarning (checking): Data.Map's `Semigroup` instance is now unbiased and differs from the left-biased instance defined in PureScript releases <= 0.13.x.
• deno@0.0.5/src/Deno/Dotenv.purs:41:20 — CustomWarning (checking) × 2: Data.Map's `Semigroup` instance is now unbiased and differs from the left-biased instance defined in PureScript releases <= 0.13.x.
• deno@0.0.5/src/Deno/Http/Request.purs:47:25 — CustomWarning (checking): Data.Map's `Semigroup` instance is now unbiased and differs from the left-biased instance defined in PureScript releases <= 0.13.x.
• literals@1.0.2/src/Literals/Null.purs:11:1 — UnparseableFFIModule (javascript): Oxc could not parse the JavaScript FFI module. Fix the invalid or unsupported JavaScript syntax; Iris treated the module as opaque and skipped export-name validation: Unexpected token
• react-basic-dom-beta@0.1.1/src/Beta/DOM.purs:33:31 — DuplicateImport (indexing): Import list contains multiple references to 'Proxy'
• sparse-polynomials@3.0.1/src/Data/Sparse/Polynomial.purs:1048:1 — MissingPatterns (checking) × 2: Pattern match is not exhaustive. Missing: _
• text-formatting@0.1.0/src/Data/Text/Format/Dodo/Printer.purs:64:25 — CustomWarning (checking) × 23: Debug function usage
• trivial-unfold@0.5.0/src/Data/Unfoldable1/Trivial1.purs:150:17 — MissingPatterns (checking): Pattern match is not exhaustive. Missing: Right _
• xterm@1.0.0/src/XTerm/UnicodeHandling.purs:15:1 — UnparseableFFIModule (javascript) × 2: Oxc could not parse the JavaScript FFI module. Fix the invalid or unsupported JavaScript syntax; Iris treated the module as opaque and skipped export-name validation: Expected a semicolon or an implicit semicolon after a statement, but found none
• yoga-react-dom@2.0.1/src/Yoga/React/DOM.purs:34:31 — DuplicateImport (indexing): Import list contains multiple references to 'Proxy'
• yoga-tree-utils@1.0.0/src/Yoga/Tree/Extended/Path.purs:20:72 — DuplicateImport (indexing): Import list contains multiple references to 'snoc'

View workflow run