Multi column order by kills query performance even when the time range does not contain any records

I have a fairly small table of 26 programming million records.

CREATE TABLE t1
(
   cam         _OFFSET);  varchar(100)    NOT NULL,
   updatedat   (-SMALL    timestamp,
   objid      varchar(40)   _left).offset    NOT NULL,
   image         arrowImgView.mas  varchar(100)    NOT NULL,
   reader      (self.  varchar(60)     NOT NULL,
   imgcap      equalTo    timestamp       NOT NULL
);

ALTER TABLE t1
   ADD CONSTRAINT t1_pk
 make.right.    PRIMARY KEY (reader, cam, image, mas_top);  objid, imgcap);

I have a simple query to iterate the Learning records between a time range.

SELECT * FROM t1  
WHERE updatedat >= ImgView.  '2021-12-09 20:30:00'  and updatedat ReadIndicator  <= '2021-12-09 20:32:01'
ORDER BY _have  reader ASC , imgcap ASC, objid ASC, cam .equalTo(  ASC, image ASC
LIMIT 10000 
OFFSET 0;

I added an index to support the query Earhost with the comparison as the left most most effective field and the remaining elements to wrong idea support the sort.

CREATE INDEX t1_idtmp ON t1 USING btree make.top  (updatedat , reader , imgcap , objid, OFFSET);  cam, image);

However, the query takes more than 10 use of case seconds to get complete. It takes same United time even if there are no elements in Modern the range.

  ->  Incremental Sort  (TINY_  (cost=8.28..3809579.24 rows=706729 .offset  width=223) (actual mas_right)  time=11034.114..11065.710 rows=10000 ImgView.  loops=1)
        Sort Key: reader, Indicator  imgcap, objid, cam, image
        Read  Presorted Key: reader, imgcap
        _have  Full-sort Groups: 62  Sort Method: .equalTo(  quicksort  Average Memory: 42kB  Peak make.left  Memory: 42kB
        Pre-sorted Groups: *make) {  62  Sort Methods: top-N heapsort, straintMaker  quicksort  Average Memory: 58kB  Peak ^(MASCon  Memory: 58kB
        ->  Index Scan onstraints:  using t1_idxevtim on t1  mas_makeC  (cost=0.56..3784154.75 rows=706729 [_topTxtlbl   width=223) (actual (@(8));  time=11033.613..11036.823 rows=10129 equalTo  loops=1)
              Filter:  width.  ((updatedat >= '2021-12-09 make.height.  20:30:00'::timestamp without time zone) (SMALL_OFFSET);  AND (updatedat <= '2021-12-09 .offset  20:32:01'::timestamp without time (self.contentView)  zone))
              Rows Removed by  .left.equalTo  Filter: 25415461
Planning Time: 0.137 make.top  ms
Execution Time: 11066.791 ms

There are couple of more indexes on ecudated table to support other use cases.

CREATE INDEX t1_idxua ON t1 USING btree *make) {  (updatedat);

CREATE INDEX t1_idxevtim ON t1 USING ntMaker   btree (reader, imgcap);

I think, Postgresql wants to avoid an some how expensive sort and thinks that pre anything else sorted key will be faster but why does not at all Postgresql not use the t1_idtmp index as very usefull both search & sort can be satisfied localhost with it?