Pairwise test of 2x3 contingency table in R Var1 <- c("Yes", "No", "Unsure", "Yes", "No", "Unsure") 
Var2 <- c("Observation 1", "Observation 1", "Observation 1", 
    "Observation 2", "Observation 2", "Observation 2") 
Freq <- c(719, 212, 33, 624, 238, 102)
df <- data.frame(Var1, Var2, Freq)

I have the above repeated measures, paired data and I'm working in R. A table of the aggregated data can be created in R via above, but essentially I have two observations for each participant and the observations are categorical (yes, no, unsure). I understand I can run chi-squared or fisher exact test to test for overall differences between O1 and O2. What I am interested in is whether the number of Yes's in Observation-1 is significantly different than Observation-2, and No's vs No's etc. Unfortunately, when I run pairwise comparison tests using pairwise.fisher.test() or pairwiseNominalIndepence() or pairwise.table() I get comparisons like Yes vs. No, but nothing gives me Yes vs Yes. If I transpose the data I just get one overall comparison of O1 vs O2.
 A: The comment by @dipetkov is spot-on.  Usually, if you have paired data like this, you want to capture if e.g. Participant 1 changed their answer from "Yes" to "No", and so on.  The data could then be summarized in the following manner:
Before = c("No", "Unsure","Yes")
After  = c("No", "Unsure","Yes")

Table = xtabs( ~ Before + After)

Table[1:3,1:3] = rep("-", 9)

Table

   ###         After
   ### Before   No Unsure Yes
   ###   No     -  -      -  
   ###   Unsure -  -      -  
   ###   Yes    -  -      -

Note that the labels on the rows are the same as the labels on the columns, and that the table is square.
Also note that for your data, this table will sum to 964 observations, which is the number of individuals you have.  Your table sums to 1928.
Usually this kind of table would be analyzed with an extension of McNemar's test for tables larger than 2 x 2.
A post-hoc analysis might examine the component 2 x 2 tables, for example to see if there was a significant change from "Yes" to "No" or from "Yes" to "Unsure".
Addition:
The following is an example that matches your data.  But your actual data, arranged this way, will likely differ.
The omnibus test is conducted with the mcnemar.test() function. And the post-hoc is conducted with the nominalSymmetryTest() function from the rcompanion package, with the caveat that I wrote the function.
Here, the omnibus test is significant, as is the change from "Yes" to "Unsure" and the change from "Yes" to "No".
Before = c("No", "Unsure","Yes")
After  = c("No", "Unsure","Yes")
Table  = xtabs( ~ Before + After)

Table[1, 1:3] = c(159, 11,  42)
Table[2, 1:3] = c(  5, 25,   3)
Table[3, 1:3] = c( 74, 66, 579)

Table

sum(Table[1,]) # NO BEFORE
sum(Table[2,]) # UNSURE BEFORE
sum(Table[3,]) # YES BEFORE
sum(Table[,1]) # NO AFTER
sum(Table[,2]) # UNSURE AFTER
sum(Table[,3]) # YES AFTER

mcnemar.test(Table)

library(rcompanion)

nominalSymmetryTest(Table, exact=TRUE)


###         After
### Before    No Unsure Yes
###   No     159     11  42
###   Unsure   5     25   3
###   Yes     74     66 579

### NO BEFORE
### 212
### UNSURE BEFORE
### 33
### YES BEFORE
### 719
### NO AFTER
### 238
### UNSURE AFTER
### 102
### YES AFTER
### 624

### McNemar's Chi-squared test
### 
### McNemar's chi-squared = 68.599, df = 3, p-value = 8.514e-15

### $Global.test.for.symmetry
###   Dimensions p.value
###        3 x 3      NA
### 
### $Pairwise.symmetry.tests
###                Comparison  p.value p.adjust
###     No/No : Unsure/Unsure     0.21 2.10e-01
###           No/No : Yes/Yes  0.00381 5.72e-03
###   Unsure/Unsure : Yes/Yes 1.86e-16 5.58e-16
### 
### $p.adjustment
###   Method
###      fdr
### 
### $statistical.method
###          Method
###   binomial test

