You're trying to fit a regression model with a time trend, and you've generated the trend variable for the training period, but not for the test period. In other words, the forecast() function doesn't know what the values of t are in the future unless you tell it.
You could fix the problem by including a newdata argument in the call to forecast() which would include the future values of the t variable.
But a simpler fix is to use the built-in trend variable which is recognized by both tslm() and forecast() as a linear time trend and the values are generated for you.
library(forecast)
arrival_ts <- ts(rnorm(1200), start= c(1921,04), end=c(2021, 05), frequency = 12)
lm1 <- tslm(arrival_ts ~ trend)
tsforecast <- forecast(lm1, h=50)
Created on 2021-09-07 by the reprex package (v2.0.1)