Build hygiene: unrendered bcf-0001 root + invalid XML on titles containing '<'

[bmorphism]

Two build-hygiene findings surfaced while building the forest from current branches (forester 4.3.1) and validating the output with an independent strict XML parser. Both are pre-existing on main (not introduced by any open PR) and are stable across PR #25 and #30.

1. Forest root bcf-0001 is never rendered → dangling transclusions

forester build emits no output/bcf-0001.xml, yet bcf-0001 is referenced via \transclude (e.g. from mat-0001 and the generated index). This is a genuine H¹/referential-integrity hole: the root of the forest has no page.

Evidence (identical in #25 and #30 builds):

transclude  mat-0001 -> bcf-0001   (target not in rendered output)
transclude  index    -> bcf-0001   (target not in rendered output)

Suggested fix: ensure the root tree is rendered (give bcf-0001 a body / confirm it isn't filtered out of output/), or point the root config at a tree that is rendered.

2. Titles containing < produce invalid XML

When a tree title contains a literal <, forester escapes the element text correctly (<) but leaves the text="…" attribute unescaped, yielding non-well-formed XML. A strict parser (Python xml.etree) rejects these; ~0.7% of machine-generated unstable-* trees fail this way.

Example (unstable-17488.xml):

<fr:title text="Immediate (Distance < 2.0)">Immediate (Distance &lt; 2.0)</fr:title>

The raw < in text="…" is invalid. Authored bcf/bci/horse trees are unaffected (none currently have < in titles): 88/88 (PR #25) and 103/103 (PR #30) core trees are well-formed. This is a latent trap — any future title with < will emit invalid XML at build time.

Suggested fix: escape </&/" in the text= attribute (upstream forester), or a post-build sanitizer pass over output/*.xml.

Reproduce

./forester build bci-forest.toml            # forester 4.3.1
# 1) ls output/bcf-0001.xml  -> missing
# 2) python3 -c "import xml.etree.ElementTree as ET,glob; \
#      [print('BAD',f) for f in glob.glob('output/*.xml') \
#       if (lambda: (ET.parse(f),False)[1] if True else None)]" 2>&1 | head

Context: found during a functorial-diff review of #25/#30 (the dangling cocycle is conserved across both — neither PR fixes or worsens it).