Increase read buffer to reduce total syscalls

[heshpdx]

This patch reduces read syscalls by 96%. strace showed that we were doing small incremental reads of large files:

openat(AT_FDCWD, "stratixiv_arch.timing.xml", O_RDONLY) = 5
read(5, "<architecture>\n  <models>\n    <!"..., 8191) = 8191
read(5, "sink_ports=\"eccstatus portbdatao"..., 8191) = 8191
read(5, "rtbdataout\" clock=\"clk_portbout\""..., 8191) = 8191
read(5, "ortbdataout\"/>\n        <port nam"..., 8191) = 8191
read(5, "ombinational_sink_ports=\"eccstat"..., 8191) = 8191
read(5, "nal_sink_ports=\"eccstatus portbd"..., 8191) = 8191
...

This might have been fine 20 years ago, but we can modernize it. Increasing the buffer size reduces the total number of syscalls, which helps in scale-out scenarios with many copies of vpr running on the same system. With this patch, strace looks like:

openat(AT_FDCWD, "stratixiv_arch.timing.xml", O_RDONLY) = 5
read(5, "<architecture>\n  <models>\n    <!"..., 1048576) = 1048576

Another win comes with in libblifparse, which is reading in large blif files. I was able to modify my generated yacc output, however I cannot figure out the correct way to capture this upstream in the build:

--- a/src/vtr-vpr/libs/EXTERNAL/libblifparse/src/blif_lexer.gen.cpp
+++ b/src/vtr-vpr/libs/EXTERNAL/libblifparse/src/blif_lexer.gen.cpp
@@ -971,12 +971,7 @@ static int input ( yyscan_t yyscanner );
 
 /* Amount of stuff to slurp up with each read. */
 #ifndef YY_READ_BUF_SIZE
-#ifdef __ia64__
-/* On IA-64, the buffer size is 16k, not 8k */
-#define YY_READ_BUF_SIZE 16384
-#else
-#define YY_READ_BUF_SIZE 8192
-#endif /* __ia64__ */
+#define YY_READ_BUF_SIZE 1048576
 #endif

Just set all the buffer sizes to 1MB, not sure why ia64 has to be different.