Fix Ghidra comment read fidelity

[amattas]

Problem

Ghidra stores several comment slots for each code unit: plate, pre, EOL, post, and repeatable comments. The current declib Ghidra reader only looks at a subset of those slots and collapses multiple comments at the same address by overwriting earlier text. In practice this loses comments when round-tripping state from Ghidra into clients such as BinSync.

Two concrete failure modes this fixes:

• EOL and PRE comments at the same address are not both preserved.
• PLATE, POST, and REPEATABLE comments are not represented when reading comments back from Ghidra.

Solution

• Read all populated Ghidra comment slots in Ghidra display order.
• Preserve multiple comments at one address by joining them with explicit slot labels.
• Set decompiled=True only when the merged comment is purely pseudocode/PRE text; mixed disassembly+pseudocode comments remain disassembly-side comments.
• Extend the existing decompiler tests with eight parameterized subTest cases covering every singleton slot, empty input, mixed classification, and full display ordering.
• Exercise all cases in one multi-code-unit traversal and preserve addr/func_addr assertions.

Verification

• TEST_BINARIES_DIR=/tmp/declib-empty-test-binaries python -m pytest tests/test_decompilers.py::TestHeadlessInterfaces::test_ghidra_comment_slots -q -> 1 passed, 8 subtests passed
• PYTHONPYCACHEPREFIX=/tmp/declib-pr199-pyc python -m compileall -q declib tests
• git diff --check
• Mutation checks confirmed failures for incorrect func_addr, last-slot-only classification, missing empty-comment filtering, first-code-unit-only traversal, and missing Java-string conversion.

Notes

This PR is independent of the BinSync PRs. It only changes how declib reads comments from Ghidra. The focused test uses an empty TEST_BINARIES_DIR only to satisfy the existing module-level fixture-directory check; the test itself uses mocks and does not consume a binary fixture.

[amattas]

Updated the test layout as requested: removed the new tests/test_ghidra_comments.py file and consolidated its coverage into parameterized subTest cases in the existing tests/test_decompilers.py. The focused test now reports 1 test and 4 passing subtests.

[amattas]

Follow-up test hardening pushed. The existing-file test now has 8 subTest cases covering empty input, all five singleton slots, mixed PLATE/PRE classification, and all-slot ordering in one multi-code-unit traversal. It also restores addr/func_addr assertions and verifies the exact Ghidra traversal calls. Mutation checks killed the previously surviving regressions.

[mahaloz]

@amattas please rebase on main (your other PR), so the conflicts get resolved

[amattas]

@mahaloz Donezo on the rebase.