Document channel width sources and trust model

[NewGraphEnvironment]

Problem

fl_valley_confine() uses two independent channel width sources from different regressions, and this comes up repeatedly when users ask whether to trust the output.

The two width sources

Source	Used for	Formula	Origin
bcfishpass channel_width	channel_buffer polygon (DEM correction)	exp(0.307) * (area * precip / 100000) ^ 0.458	Poisson Consulting 2021b
VCA bankfull regression	Flood depth modelling (fl_flood_surface)	(area ^ 0.280) * 0.196 * (precip ^ 0.355)	Hall 2007 / USDA VCA

These are structurally different — Poisson uses a single exponent on a discharge proxy (area * precip), while the VCA uses independent exponents for area (0.280) and precip (0.355). They will give different widths for the same stream.

Additional complications

• Order 1 streams have NA channel_width in bcfishpass — flooded skips these for the buffer but includes them via rasterization and flood model
• channel_buffer = FALSE disables only the bcfishpass-derived buffer; the VCA flood model still runs with its own regression
• Users may not realize the flood extent and channel buffer come from different models

Proposed Solution

Document this in fl_valley_confine() roxygen @details and/or the valley-confinement vignette:

• The two width sources, where each is used, and why they differ
• When channel_buffer = FALSE is appropriate (don't trust bcfishpass widths, or DEM resolution already captures the channel)
• NA channel_width handling for order 1 streams
• Pointer to fresh for width estimation/fill (Fill NA channel_width for small order streams fresh#28, Estimate channel width from bankfull regression fresh#29)

References

• bcfishpass/model/03_habitat_lateral/valley_confinement.md (line 30)
• fwapg/extras/channel_width/sql/channel_width_modelled.sql
• bcfishpass/docs/03_model_habitat_linear.md

Relates to NewGraphEnvironment/fresh#28
Relates to NewGraphEnvironment/fresh#29
Relates to #21