Fix iplot parsing for IV sunab terms

[kmfrick]

iplot() errors on valid IV models that use sunab(..., no_agg = TRUE) in an interacted endogenous term, even though the coefficient names are valid and plottable.

On the econometrics side, you may argue that Sun & Abraham do not cover fuzzy designs, so this is working as intended as we don't know what the Wald-DiD estimator yields under staggered treatments (do we? maybe under homogeneous effects but staggered treatments?). But then I think it should at least output a clearer error like the "without a doubt, your model is misspecified" when you have only variables that are colinear with the FEs. I may do that if you want.

Repro

library(data.table)
library(fixest)

set.seed(20260213L)
n_id <- 400L
never_cohort <- 100L

dt <- CJ(id = 1:n_id, rel_time = -2:2)
dt[, high_growth := as.integer(id <= n_id / 2L)]
dt[, cohort := fifelse(high_growth == 1L, 0L, never_cohort)]
dt[, x := rnorm(.N)]
dt[, id_fe := rnorm(n_id)[id]]
dt[, y := 0.08 * x * (rel_time >= 0L) * high_growth + id_fe + rnorm(.N, sd = 0.25)]

iv_model <- feols(
  y ~ 1 |
    x:sunab(cohort, rel_time, ref.p = -1, ref.c = never_cohort, no_agg = TRUE) ~
    sunab(cohort, rel_time, ref.p = -1, ref.c = never_cohort, no_agg = TRUE),
  data = dt
)

iplot(iv_model)

Actual behavior

iplot() throws:

In iplot(), no valid variable created with i() found...

Expected behavior

iplot() should plot the IV sunab coefficient

Root cause

In IV stage 2, iplot() used broad fit_ prefix stripping, which would fail to recover the correct i()/sunab term names.

I think there was an issue in the past with this (which maybe I reported myself? not sure), so I was extra cautious in stripping only actual endogenous variables that were fitted in the first stage.

Fix

• Restrict renaming to known fitted endogenous names via object$iv_endo_names_fit.
• Strip only the fitted-endo interaction prefix (fit_<endo>:) where appropriate.
• Keep a fallback path for legacy objects without iv_endo_names_fit.

Tests

Added regression coverage in tests/fixest_tests.R:

1. IV + sunab(..., no_agg = TRUE) interacted fitted regressor now works with iplot().
2. IV with an exogenous i() variable named with fit_ prefix (e.g. fit_user) still works, preventing over-matching regressions.

Verification

• Targeted test chunk (Interact) passes.
• Repro script now reports IV iplot: OK.

Disclaimer

I noticed the error while writing code for a WP and thought it'd be pretty easy to fix. I had Codex write the correction, and it just changed the regex, but I was suspicious so I probed it a bit more and found the issue with variables that start with "fit_", so I had it make the code robust to that as well. Tests pass, I have experience with R and the code smells okay to me, but let me know if you want me to check more thoroughly something specific.