Modify the plugin.ObjFile interface by replacing the "Base() uint64" method
with an "ObjAddr(addr uint64) (uint64, error)" method. The latter computes
the objdump address corresponding to the given runtime address.

The change removes direct access to the base value, but enables us to delay
computing the base until it is actually needed.

…623)

We can use the sample's offset into the mapping to differentiate between small
segments that are in the same file page when associating a segment with the
mapping.

…onfigurations run if one matrix configuration fails. (#627)

* Source listing handles missing binaries and addresses.

Some profiles can contain locations without an address, or reference
binaries/libraries that are not available. Treat these better when
producing source listings by using any source information present
directly in the profile.Location.

Details:

If a location does not have an address, create a synthetic address
that does not fall in a known mapping.

If a location references an object file that is not found, use any
file/line information baked directly in the location.

Tweaked generated html to clicking for source lines that have no
inlining/assembly to display.

* Panic on attempt to synthesize address for Loc with non-zero addr.

* feat: add link to download profile to the UI

* use space

* use space

* add semicolon

…635)

Changed ProgramHeadersForMapping to get in a slice of program headers instead of
an elf.File object.

Moved the logic for matching headers by file offset to the elfexec package.

Removed testelf package, as it didn't have that much reuse. Instead, defined the
relevant test headers in the individual test files.

Other refactoring inside binutils to simplify code.

No functional changes.

list, weblist, disasm, and peek all take a regexp argument that acts as
a filter.

Currently if this filter results in no matches we simply generate an
empty report, which can be confusing to users. Instead, explicitly treat
this as an error condition that is clearly reported to users.

Fixes #629

* Remove the strict test on segment memory size.

It's not precise in some corner cases. It assumes a mapping includes at most one
segment, but the loader may map memory for the entire binary at first, and under
the right circumstances, this test can select the wrong segment. The filtering
by sample offset works more reliably.

Also, the strict test on segment memory size is less effective after adding
support for matching split mappings to segments.

* Add support for matching split mappings to segments.

Update ProgramHeadersForMapping to match a mapping to any segment with
overlapping file offsets, except if:
- if the mapping starts before the page aligned offset associated with a segment.
  There is no known instance where the loader would load the content of a segment
  to include preceding pages of the file that don't include any segment content.
  The mapping must be associated with an earlier segment in such cases.
- if the mappings includes the last page of the segment, but not the full
  segment, and it includes additional pages after the segment end. It's more
  likely that the mapping is associated with the following segment in such cases.