multistage random sampling in r not based on rows, but based on group and unique values in another column

I want to carry out multistage random sampling on a master dataframe, in order to compare results of different sample sizes. Different examples that I have seen on stackoverflow have carried out sampling based on number of rows, but that is not exactly the expectation in this task. Overview of the master dataframe is this:

glimpse(df)
Rows: 23,404
Columns: 4
$ variety <dbl> 330, 330, 330, 330, 330, 330, 251, 251, 251, 251, 251, 251, 312, 312, 312, 312, 312, 312, 312, 31...
$ plantno <dbl> 13, 13, 13, 13, 13, 13, 19, 19, 19, 19, 19, 19, 23, 23, 23, 23, 23, 23, 23, 23, 29, 29, 29, 29, 2...
$ point   <chr> "L.0000", "L.0001", "R.0000", "R.0001", "R.0002", "R.0003", "L.0000", "L.0001", "L.0002", "R.0000...
$ length  <dbl> 31.9868410, 25.7816640, 18.4944923, 19.5613560, 10.7766413, 10.1496060, 10.7168411, 8.5189431, 10...

The sampling are done in two stages:

• 1st stage - sample plants per variety (this is not equivalent to rows). Sample sizes are 5 10 15 20 25 30 35. For instance, sampling 5 plants per variety will not mean sampling five rows but sampling unique five plantno, which in real sense goes along with their replicates. The picture below shows a correct example of sampling after stage 1. It shows that under variety 201, five plants 15 10 18 17 36 from plantno column were sampled.

• 2nd stage - sample point per plantno. Sample sizes are 2 4 6 8 10. The picture below shows a correct example of sampling after stage 2. For example, sampling 2 points per plantno shows that under variety 201, plantno 10 & 15 were sampled from the 1st stage, while for the 2nd stage, L.0002 & R.0004 were sampled from plantno 10, and L.0002 sampled from plantno 15. Stage 2 is actually the final result. Each sampling cycle is stored in a list which are later joined together.

The attempt I made according to the code below did not give me expected result. The result seem to be filled with duplicates, and not according to the table shown above. The . Any help will be appreciated.

library(purrr)
library(tidyr)
library(rio)
library(tidyverse)

# import data from the web to your local computer
myurl <- "https://drive.google.com/file/d/1ym_s5KeqcQ9yZFBK2nO6FGhC_WYVUv8x/view?usp=sharing"
download.file(url=myurl, destfile="df_test1.xlsx", mode="wb")

# import multiple excel workbook using rio library
df <- import_list(file.choose(), setclass = "tbl")
df <- df[[1]]
glimpse(df)

## generate sequence of sample sizes for tree per family
J <- seq(5,35,5)
# l1 <- vector(length = length(J), mode = "list")

## generate sequence of points per tree
k <- seq(2,10,2)
l2 <- vector(length = length(k), mode = "list")
l3 <- vector(length = length(J), mode = "list")

for (i in seq_along(J)){    # you know of seq_along() right?
  l1 = df %>%
    group_by(plantno, variety) %>%
    # distinct %>%
    sample_n(J[i], replace = T)
  # 
  # l1 <- l1 %>% left_join(df)
  
  sampleSize_variety <- rep(J[i], nrow(l1))
  
  l1$sampleSize_variety <- sampleSize_variety
  
  for (j in seq_along(k)){  
    l4 = l1 %>%
      group_by(variety, plantno) %>% 
      sample_n(k[j], replace = T) 
    
    sampleSize_pointPerPlant <- rep(k[j], nrow(l4))
    
    l4$sampleSize_pointPerPlant <- sampleSize_pointPerPlant
    
    l2[[j]] = l4
  }
  l3[[i]] <- l2
}

# bind tables together - tables are in nested list
merge_table <-  transpose(l3) %>% map(bind_rows) 
merge_table1 <- bind_rows(merge_table)
merge_table1