utils_c.interp.interp_conserve_c lose flux at particular wavelength grid

[Fmajor]

from grizli.utils_c import interp
iy = interps.interp_conserve_c(ix, x, y, integrate=0, left=0, right=0)

This function may lose flux at boundary points at some particular wavelength grid

the origin x,y is from the sensitivity of beam B
the ix is the lam calculate with axeconf.get_beam_trace

In the below figures, I do interp_conserve_c using three methods
jin method is my method, xin method is from Xin Zhang

In this wave grid, all three method are consistent

However, In this wave grid, the grizli method seems to lose flux (blue dashed line, in the left bottom point)

[Fmajor]

grizli-interp-function-debug-data.fits.txt

here is the data for you to debug, rename it to fits type.

Filename: ./grizli-interp-function-debug-data.fits
No.    Name      Ver    Type      Cards   Dimensions   Format
  0  PRIMARY       1 PrimaryHDU       4   ()
  1                1 BinTableHDU     29   5301R x 3C   ['D', 'D', 'D']
  2                1 BinTableHDU     12   60R x 2C   ['D', 'D']

import astropy.io.fits as fits
from astropy.table import Table
dobj = fits.open('./grizli-interp-function-debug-data.fits')
sens = Table.read(dobj[1])
lams = Table.read(dobj[2])
x = sens['WAVELENGTH'].data.byteswap().newbyteorder()
y = sens['SENSITIVITY'].data.byteswap().newbyteorder()
lam0 = lams['lam0'].data.byteswap().newbyteorder()
lam1 = lams['lam1'].data.byteswap().newbyteorder()

lam0 is the wave grid in figure2 (the bad one)
lam1 is the wave grid in figure1 (the good one)