fix(dev): correct ITS primers to gITS7/ITS4; add the ITS case to the prep page

[cjfields]

dev/cutadapt_ITS.sh was a copy of the V3–V4 script with ITS1f/ITS2 pasted in, taken on the strength of the NCBI annotation Design: ITS1FI2-ITS2 (ITS2). Verified against the reads (two samples, 50k reads, 5' end), neither named primer is present:

primer	R1	R2
gITS7 GTGARTCATCGARTCTTTG	92.5% (offset 0, no pad)	0%
ITS4 TCCTCCGCTTATTGATATGC	0%	90.9% (offset 4, constant GATA pad)
ITS1F / ITS1FI2 / ITS1 / ITS1F_KYO2	0%	0%
ITS86F	49.3%	0%

ITS86F is the non-degenerate form of gITS7 and matches only the subset where the degenerate base is A — confirmed by base composition at that position (53.4% A / 46.6% G, i.e. real degenerate synthesis).

As written the script would have stripped nothing. Corrected, it removes both primers to 0.00% residual at 97.6% pair retention, leaving R1 median 231 nt (250−19) and R2 226 nt (250−24), exactly as predicted.

The inverse failure

gITS7+ITS4 amplifies the ITS2 region, so here the region label is right and the primer names are wrong — the opposite of the 16S arm in the same deposit. The trap is that "ITS2" names both a region and a primer: ITS1FI2 + the ITS2 primer would amplify ITS1, so the annotation contradicts its own region label. Reading either half as authoritative sends you somewhere different.

Two ITS-specific hazards now documented

• ITS length varies by hundreds of nt across taxa, so any --trunc-len is a taxonomic filter, not a quality one — it systematically deletes every fungus above the cutoff. Same asymmetry at the merge step: long-ITS taxa may fail to overlap and drop out, so an ITS merge rate is not taxonomically neutral the way a 16S one is.
• Short amplicons read through into the far primer, so -a/-A are load-bearing here rather than belt-and-braces.

Neither hazard exists on a fixed-insert 16S amplicon, so 16S habits do not transfer.

Also worth noting: the offsets here are constant (R1 at 0, R2 at 4), so a fixed --trim-left would work on this arm — unlike the 16S arm. Spacer layout is a property of a prep, not a platform, and has to be measured per dataset rather than assumed in either direction. The script's diagnostic now prints offset distributions for both primers and says which reading implies which.

Adds a section to docs/findings/reading-the-prep.md recording the ITS case as the second example.