Add user guide page 18: reconstruction from homology

[edkerk]

Adds user guide page 18, Reconstruction from homology. Unlike the ftINIT
page, nothing here is skipped or precomputed: both tabs run a real
bidirectional BLASTP and build a real draft model, in about fifteen
seconds each.

The template is smallYeast as sce; the organism being reconstructed is
H. polymorpha. That puts the page on the same ground as the hanpo-GEM protocol
without repeating it — the protocol does the full-scale version.

Both toolboxes agree throughout

	MATLAB	Python
BLAST hits	159 + 178	337
draft	37 rxns, 49 mets, 54 genes	37 rxns, 49 mets, 54 genes
with maxE=1e-100, minLen=250	32 rxns, 46 genes	32 rxns, 46 genes

They call the same BLAST+ binaries with the same parameters, so this is a real
cross-check rather than a coincidence.

18.4 is the point of the page

objective set: 0
exchange reactions: 0
reactions that can carry flux: 0 of 37

Not one reaction in the draft can carry flux. That is the normal state of a
homology draft — a set of claims about which reactions the organism probably
has, and nothing else — and it is why the rest of the guide exists. The section
hands off to media, gap-filling and tasks.

Also documented: minLen's default changed from 200 to 100 (measured against
KEGG and OMA orthology), so an older reconstruction script now produces a
different model unless the value is set explicitly.

New data files

file	size	provenance
hanpo.faa	2.4 MB	copied unchanged from the hanpo-GEM submodule
sce-template.faa	31 KB	the 61 S. cerevisiae sequences that are exactly smallYeast's genes

Cutting the template proteome down to the model's own genes is what keeps the
example to seconds rather than hours; a template protein that no reaction uses
could not have contributed to the draft anyway. Provenance and the regeneration
command are in docs/data/README.md.

Two things to watch on CI, both new to this page: the Python job downloads BLAST+
from raven-data on first use (RAVEN ships its own), and the harness copies
docs/data into every page directory, so the 2.4 MB proteome is copied per page
for the MATLAB run.

Checks

• 10 examples, both languages, all passing
• check_names.py clean (six argument names added to the allowlist)
• site builds

The guide is now 18 pages, 185 examples, no skips.