flamegraph panics on empty lines with --reverse set

[mrob95]

Found while I was investigating #237

Repro:

• Change tests/data/flamegraph/bad-lines/bad-lines.txt to the following (adding a blank line):

cksum;_start;__libc_start_main;main;cksum 31
cksum;cksum 6
THIS IS A BAD LINE
cksum;cksum;__GI___fread_unlocked;_IO_file_xsgetn;_IO_file_read;entry_SYSCALL_64_fastpath_[k];sys_read_[k];vfs_read_[k];__vfs_read_[k];ext4_file_read_iter_[k] 1
cksum;main;cksum 19
THIS IS A BAD FRACTIONAL LINE 12V.43
noploop;[unknown] 2
noploop;main 274

noploop;[unknown] 2

• Run cargo test flamegraph_should_warn_about_bad_input_lines_with_reversed_stack_ordering

Failure:

---- flamegraph_should_warn_about_bad_input_lines_with_reversed_stack_ordering stdout ----
thread 'flamegraph_should_warn_about_bad_input_lines_with_reversed_stack_ordering' panicked at 'attempt to subtract with overflow', /home/mike/inferno/src/flamegraph/mod.rs:415:60
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    flamegraph_should_warn_about_bad_input_lines_with_reversed_stack_ordering

The issue is here: https://github.com/jonhoo/inferno/blob/master/src/flamegraph/mod.rs#L411

samples_idx is 0 when we have an empty line, and we are trying to subtract from it.

A simple fix would be to add a check for line.is_empty() at the top of that loop. Alternatively, it seems like the reason we are subtracting 1 from sample_idx is to remove a trailing space from the line, so replacing with &line[..samples_idx].trim_end() would fix it.