Fix for #7363.

What's happening

On a repo with core.symlinks=false (a filesystem without symlink
support, or a caller that asks libgit2 to materialize symlinks as
regular files via git_futils_fake_symlink in blob_content_to_link),
a GIT_CHECKOUT_SAFE that only needs to delete an unmodified
symlink entry fails with GIT_ECONFLICT / "N conflicts prevent
checkout" — while git_status on the same workdir reports fully
clean, and the git CLI does the same checkout without complaint.

The status/diff side already handles this: diff_file_content__load_workdir
looks up GIT_CONFIGMAP_SYMLINKS and treats a link entry as file
content when symlinks are unsupported. Only checkout.c's dirtiness
check never consulted that config — is_filemode_changed normalizes
S_IFLNK only for core.filemode=false, and the workdir plain file
that libgit2 itself wrote is compared against an S_IFLNK index/base
entry as if the mode had changed.

The same phantom conflict hits GIT_DELTA_MODIFIED (symlink whose
target text changed between branches) and GIT_DELTA_TYPECHANGE
(symlink ↔ blob/tree) — anything that flows through
checkout_is_workdir_modified with a link on either side.

The change

Add a respect_symlinks parameter to is_filemode_changed and pass
data->can_symlink at the three call sites in
checkout_is_workdir_modified. When !respect_symlinks, both sides
are normalized to GIT_FILEMODE_BLOB before comparison. The two
branches (respect_filemode and respect_symlinks) are independent —
respect_filemode=false still handles the executable bit as before,
and the new block only touches S_IFLNK.

+15 / −4, src/libgit2/checkout.c only.

How it was verified

Ran the reporter's self-contained C reproducer verbatim, against a
Debug build of main at 0551dfd:

• Baseline: workdir status entries: 0 (clean),
  SAFE checkout B: rc=-13 message="1 conflict prevents checkout" —
  matches the report exactly.
• With this change: workdir status entries: 0 (clean),
  SAFE checkout B: rc=0 message="", and the materialized
  AGENTS.md is removed. CLAUDE.md (unchanged blob) still evaluates
  as unmodified on the same path — verified with a stderr trace at
  is_filemode_changed showing the second call returns false the same
  way it always did (a=100644 b=100644 rs=0).

Reporter

Thanks to @liz7up — the report has the syscall/API-level trace and
the exact fix direction, so this could ship as a small change with
GIT_CONFIGMAP_SYMLINKS already plumbed through checkout_data.

AI assistance

I used an AI assistant to help draft the change, the reproduction and
this description. Every line was read, checked and run by me.

• I have re-read every line and this description is in my own words.