chore(release): v1.6.3

[takeokunn]

Cuts v1.6.3. Patch level: git diff v1.6.2..HEAD adds no command, flag, JSON
field, exit code, or Nix output — every change is a fix or a text-output change,
which docs/notes/releasing.md classifies as patch.

1af7996b (parallel multi-file analysis) is fully reverted by 3868c093;
git diff 1af7996b^ 3868c093 is empty, so it contributes nothing to this
release.

Contents

• Cargo.toml / Cargo.lock: 1.6.2 → 1.6.3 (lock diff is version lines only).
• docs/src/getting-started.md, docs/src/reference/compatibility.md: pinned
  install examples updated to v1.6.3.

What ships

Emacs Lisp formatting — edit format --dialect elisp now indents special
forms, nested special forms, and buffer/conditional forms the way Emacs'
indent-region does; honours indent properties at narrow widths; keeps
escaped ? symbols intact; and no longer treats a lambda parameter named after
a special form as one. Blank lines inside progn, let, when and definition
bodies survive formatting, and Emacs Lisp now defaults to --max-blank-lines 1
(an explicit flag still overrides it).

edit normalize-quotes — --all rewrites every form the selector matches.
The flag was ignored here before: the command routed through resolve_one, so a
multi-match selector was refused as ambiguous
(packages/core/cli/src/shared.rs:283) rather than rewritten.

edit reindent-defun — a top-level definition that starts indented is moved
to column zero.

inspect capabilities — exits 0 when the consumer closes stdout early
(| head) rather than failing on the broken pipe.

Verification

nix flake check is running locally on aarch64-darwin, the platform this
repository's CI does not cover — CI fans out over x86_64-linux only. Result is
reported in a comment before merge.

[takeokunn]

Local verification on aarch64-darwin, the platform CI does not cover (ci.yml
fans out over x86_64-linux only).

nix flake check — nextest ran 15304 tests, 15303 passed. The single
failure was paredit-feature-lint-destructive-sequence cost_tests::cost_the_shipped_rule_is_local at 123465676ns against a
566612ns control, a 218x ratio on an assertion that the two stay near each
other. That crate is not in this release's diff (git diff --name-only v1.6.2..HEAD | grep -c lint-destructive-sequence → 0), and the test passes
isolated in 3.5s (cargo nextest run -p paredit-feature-lint-destructive-sequence --test-threads 1 cost_the_shipped_rule_is_local). It is a wall-clock ratio
measured under a 15k-test parallel run, not a regression. CI's Verify (nextest)
passed on this commit.

The five derivations nix listed alongside it were cancelled by the fail-fast,
not failed, so they were built separately and all pass on darwin:
treefmt (4686 files, 0 changed), treefmt-pr-check,
lint-format-integration, lint-format-integration-pr-check, package.

Release gates from docs/notes/releasing.md

nix build .#                      exit 0
./result/bin/paredit --version    paredit 1.6.3
inspect capabilities --output json  "version": "1.6.3"
cargo audit --deny warnings       exit 0, 239 crate dependencies scanned

The blank-line default, checked against the shipped binary — same source,
--max-blank-lines unset, dialect the only difference:

$ printf '(defun a () 1)\n(defun b () 2)\n' > /tmp/adj.el   # elisp
(defun a ()
  1)
(defun b ()
  2)

$ printf '(defun a () 1)\n(defun b () 2)\n' > /tmp/adj.lisp # common lisp
(defun a ()
  1)

(defun b ()
  2)

Emacs Lisp leaves the forms adjacent; every other dialect still inserts the
blank line. That is what fa6093be corrects in the flag help and the
format.blank-lines-max-consecutive schema summary, both of which still claimed
the dialect-independent behaviour.