The docs read against the graph diff

[torstei]

The release doc pass for timbersh-v0.9.0 — timbersh-v0.8.0..main, which is
timbergraph and the graph statement (#171). Reading the prose against the
code found one real defect and four gaps.

The defect

graph refuses samples that already carry a host label, because the shell
adds one itself. The message named metric, which is not bound — metrics is
— so raising the refusal raised NameError: name 'metric' is not defined from
inside the shell instead. Fixed, and the new test fails without the fix:

FAIL  a_graph_refuses_a_host_label_it_would_have_to_write_over
timberfs=#   ! name 'metric' is not defined

The gaps

• A .B request sitting mid-line in timbersh(1) is not a request: the
  rendered page carried the literal .B "graph M from SRC extracting NAME".
  Swept every man page for the same shape — this was the only one, and all six
  render with no --warnings=all output.
• --width/--height had no help text and no man entry. They are characters
  on the terminal and pixels in an image, which is worth writing down.
• The grammar \? prints — which timbersh(1) explicitly calls the authority,
  "this page deliberately does not copy it" — listed png but neither svg
  nor using. Both, and the parse error beside it, now do.
• tools/README.md documented --using under timbergraph and never
  mentioned that the graph statement has the same clause.

Packaging

Recommends: timberfs (>= 0.26.0) → >= 0.32.0. graph ... extracting runs
timberfs tally --try locally and resolves an extractor name against the
directory the timberfs package ships; both landed in 0.32.0. Still a
Recommends and still only a hint — the case that matters is a remote target,
about which a dpkg constraint says nothing, so the runtime diagnosis stays the
real guard.

⚠ Merge order: this names 0.32.0, so v0.32.0 ships first and
timbersh-v0.9.0 is tagged only after its apt-repo rebuild has finished — the
two tags must never be pushed concurrently.