[jeremiepas]

Applies the avi-1765-add-tree-sitter-grammars-php-kotlin-perl-csharp-java-imports openspec change.

The four grammars were already registered (AVI-1716). Checking each task against the spec deltas, and running graphos over a mixed-language fixture corpus, showed the languages still did not produce usable graphs.

What was broken

Gap	Effect
.pl/.pm/.psgi absent from fecCode	Perl files were never detected — the registered grammar was unreachable
PHP import forms absent from apiSurfaceTypes	use Foo\Bar; and include/require emitted nothing
C# root compilation_unit unregistered	.cs files had no module node, so their using edges had no source
extractSpecifier only read quoted literals	PHP/Perl/C#/Kotlin Import nodes had an empty label and no imports edge — forbidden by the import-resolution delta
Three dead Perl node names	use Foo; is use_no_statement, sub f {} is function_definition, quoted require 'x.pl' does not parse at ABI 11
Java resolution walked ancestors only	a src/test/java class could never resolve an import of a src/main/java class
Kotlin had no resolver	orphan Import nodes — same spec violation
Identifier-less declarations	Perl package, PHP/C# namespace, C#/Java fields all got node id "" and collided into one node per graph

What changed

• Extraction — Perl extensions in fecCode; PHP import forms and C#'s compilation_unit whitelisted; Perl node names corrected against the vendored parser.c and documented in vendor/README.md.
• Import edges — Convert.bareSpecifier reads unquoted specifiers, including the C# alias directive using A = B.C;. Java consults each ancestor's src/main/java and src/test/java, resolves import static a.b.C.m to the enclosing type, and never targets the importing file. Kotlin (LangKotlin) resolves against src/main/kotlin, then the Java roots for interop, then an external node.
• Node identity — Core.extractIdentifier learns package_name / namespace_name, plus two fallbacks keyed on the declaration type so a qualified_name or a nested identifier is never mistaken for a name where it is really a reference.

Verification

• graphos-test: 1663 examples, 1 failure, 4 pending (84 s).
• graphos over a PHP/Perl/C#/Kotlin/Java fixture corpus: every language produces symbols, 14 imports edges, 0 empty-id nodes. A src/test/java class resolves to both its src/main/java imports; a Kotlin file resolves to a .kt sibling and to a .java class via interop.

Two pre-existing issues, both unrelated to this change

1. The one test failure is the TypeScript type_identifier case. It belongs to the separate, unapplied avi-1440-ts-declaration-only-type-nodes change: that change's regression test is committed but its task 1.1 (remove type_identifier from apiSurfaceTypes) is not done. Left alone deliberately.
2. cabal test graphos-test does not terminate on main. BoundedDeltaCommunitySpec.hs:224, "delta update equals full rebuild on small random graphs (property)", spins indefinitely. Verified at 362c1fc with this branch's changes stashed — identical hang — so CI's cabal test all is blocked independently of this work. The figures above come from a run with --skip "small random graphs". This likely needs its own issue.

[jeremiepas]

Independent confirmation of pre-existing issue 2 in the description: the test job on main (run 43291562097, merge of #254) started at 19:40:35Z and was still running 2h37m later. Every other job in that run — build, fixture, haddock, studio-contract — succeeded.

So cabal test all is hung on main right now, and will hit GitHub's 6h job timeout. The cause is tests/Graphos/Domain/Graph/BoundedDeltaCommunitySpec.hs:224, "delta update equals full rebuild on small random graphs (property)", reproduced locally on bare main at 362c1fc with this branch's changes stashed.

This PR's own test job will hang the same way — it is not a signal about this change. Local result with that one property skipped: 1663 examples, 1 failure, 4 pending in 84s, the failure being the unrelated AVI-1440 TypeScript case.

Worth a separate issue; this PR does not touch community-detection code.