Remove the deprecated `field` argument from fl_valley_confine()

[NewGraphEnvironment]

Problem

0.6.0 renamed fl_valley_confine()'s field argument to a required area_field (#47) and kept
field as a deprecated alias that warns and forwards. That shim is a one-release courtesy, not a
permanent second spelling — while it exists, field = "channel_width" still runs, warning but
producing the wrong floodplain.

Remove it once the known callers have moved.

Known callers still on field =

• floodplains/scripts/floodplain_lcc/02_floodplain_model.R:148 — the production driver. Works
  today on the deprecation warning; needs a one-word edit to area_field.

Nothing inside flooded uses it: every vignette, example, test and the README moved in the same PR.

What removal looks like

• Drop field = NULL from the signature and the shim block in R/fl_valley_confine.R
• Drop @param field from the roxygen
• Drop the three shim tests in tests/testthat/test-fl_valley_confine.R
  ("deprecated field= warns and forwards", "area_field wins when both spellings are supplied", and
  the expect_no_warning negative control's counterpart)
• Note the removal in NEWS.md — it is a hard break for anyone who ignored the warning

Not covered by removal

The residual from #47 that no argument change can close: fl_valley_confine() accepts a
pre-rasterized SpatRaster and cannot inspect what quantity its cells hold. 0.6.0 warns when the
layer is named after fl_stream_rasterize()'s "channel_width" default, which catches the one
composition the package's defaults lead you into, but a raster burned from any other wrong column
is undetectable. That is stated in the roxygen and in NEWS rather than guarded, deliberately — a
magnitude guard cannot separate a small headwater basin from a channel width.

[NewGraphEnvironment]

One trap for whoever does this, surfaced in #47's review:

local_mocked_bindings is not involved, but testthat 3.1.5 stopped expect_no_warning() from
capturing deprecation warnings. The shim currently uses base warning(), so the two negative
controls in tests/testthat/test-fl_valley_confine.R ("the area_field path is silent", and the
expect_no_warning inside "area_field is not required when streams is already rasterized") really
do guard it.

Switching the shim to lifecycle::deprecate_warn() — the obvious thing to reach for while working
this issue — would silently stop those controls guarding the deprecation path, and they would stay
green. lifecycle was deliberately rejected in #47 for a related reason: it warns once per caller
by default, so the second expect_warning() in a file fails because the first consumed it.